Phylogenetic reconstruction of Phalaenopsis (Orchidaceae) using nuclear and chloroplast DNA sequence data and using Phalaenopsis as a natural system for assessing methods to reconstruct hybrid evolution in phylogenetic analyses

Two phylogenies of Phalaenopsis (Orchidaceae) are presented, one from combined chloroplast DNA data and one from a nuclear actin gene. We used these phylogenies to assess and modify the classification of Phalaenopsis and to examine several morphological characters and geographical distribution patterns. Our results support Christenson’s (2001) treatment of Phalaenopsis as a broadly defined genus that includes the species previously placed in the genera Doritis and Kingidium. Some of Christenson’s subgeneric groups needed to be recircumscribed to reflect a natural classification. We recognized four subgenera and six sections, subgenera Aphyllae, Parishianae (with sections Conspicuum, Delisiosae, Esmeralda, and Parishianae), Phalaenopsis, and Polychilos (with sections Fuscatae and Polychilos). In order to find a set of universally amplifiable, phylogenetically informative, single-copy nuclear regions, we conducted a whole genome comparison of the rice (Oryza sativa) and Arabidopsis thaliana genomes. We constructed a database of both genomes and searched for pairs of sequences using criteria we felt would ensure primers that would reliably amplify using standard PCR protocols. We tested the most promising 142 primer pairs in the lab on eighteen taxa and found four potentially informative markers in Phalaenopsis and one in Helianthus. Our results indicated that it will be difficult to find universal nuclear markers, however our database provides an important tool for finding informative nuclear markers within specific groups. The full set of primer combinations is available online at, “The Conserved Primer Pair Project,” http://aug.csres.utexas.edu:8080/cpp/index.html. We used fourteen Phalaenopsis species and seven horticultural hybrids to create a real dataset with which to test phylogenetic network reconstruction methods. We tested the performance of Neighbor-Net, implemented in SplitsTree, under four different categories of complexity: one hybrid, two independent hybrids (hybrids with no parents in common), three independent hybrids, and two non-independent hybrids (one parent was shared between hybrids). Neighbor-Net was able to predict accurately the parents of hybrids in only about half of the datasets we tested, and there were so many false positives that it was impossible to distinguish the hybrids from the species. We plan to use this dataset to test methods, such as RIATA and RGNet, when they become available.