PHYLOGENETICS OF DILLENIACEAE USING SEQUENCE DATA FROM FOUR PLASTID LOCI (rbcL, infA, rps4, rpl16 INTRON) James W. Horn1 Department of Biology, Duke University, Durham, North Carolina 27708, U.S.A., and Fairchild Tropical Botanic Garden, 11935 Old Cutler Road, Coral Gables, Florida 33156, U.S.A. Dilleniaceae are an angiosperm family consisting of 10���14 genera and 500 described species, with a pantropical distribution extending into temperate Australia. This study addresses the infrafamilial relationships of Dilleniaceae withnucleotidesequencedatafromtheplastidloci rbcL,infA, rps4,andthe rpl16 intron.Analysesofthesedatausing maximumparsimonyandBayesianmethodsresolve Tetracera,theonlypantropicalgenusinthefamily,assistertoall other Dilleniaceae. Within the clade of Dilleniaceae exclusive of Tetracera, the New World endemic genera form a cladethatissistertoacladecomposed oftheOldWorldendemicgenera. Thelatter containstwo majorsubclades:(1) a clade containing Acrotrema, Dillenia, and Schumacheria and (2) a clade containing Hibbertia and its satellite genera, Adrastaea and Pachynema, which are embedded within Hibbertia. Ancestral-state reconstructions of six morphological characters of both biological and taxonomic significance within Dilleniaceae suggest hypotheses of polarity or lability for each that differ substantially from those based on evolutionary trends. Perforation plate���type evolution within Dilleniaceae is equivocal, but the large majority of most parsimonious reconstructions suggest that simple perforation plates are plesiomorphic. Poorly organized leaf venation architecture is synapomorphic for Hibbertia s.l. Floral morphological features classically regarded as primitive, such as exceptionally large numbers of stamens (200��) and numerous carpels (more than isomerous with the corolla), are clearly derived within Dilleniaceae and are features of uncommon occurrence within the family as a whole. Both multicarpellate, synorganizedgynoeciaandmonosymmetricandroeciahavemultipleoriginswithinthefamily.Anewclassificationof Dilleniaceae is outlined, and nomenclatural changes necessitated by the phylogenetic results are provided. Keywords: Dilleniaceae, character evolution, molecular systematics, floral morphology. Online enhancement: appendix table. Introduction Dilleniaceae are a family of trees, shrubs, and lianas contain- ing 10���14 genera and 500 species, with a pantropical distri- bution that extends into temperate Australia (fig. 1 Horn 2007). Although remarkably diverse, especially in habit and floral structure, the family is readily diagnosed by a unique suite of morphological features. Of these, vegetative characters are the most reliable means of identifying the family over much of its geographic range. The combination of scabrous leaves and primary stems leaf venation with more or less straight, parallel secondaries and scalariform tertiaries petioles with a broad insertion at the stem and orangish inner or outer bark that exfoliates thin plates or strips is diagnostic. Floral characteristics that distinguish the family include uniformly persistent and typically accrescent calyces apopetalous, cadu- cous corollas polymerous, marcescent, or tardily deciduous androecia free stylodia and arillate seeds. Thus identifiable, Dilleniaceae have been considered a natural grouping within angiosperms whose circumscription, apart from the inclusion of Actinidiaceae by several early authors (e.g., Baillon 1865), has remained stable (de Candolle 1824 Gilg and Werdermann 1925 Hutchinson 1964 Dickison 1970b Takhtajan 1997). Molecular systematic studies to date, in accordance with this perception, strongly suggest the monophyly of Dilleniaceae, but they sample a narrow range of diversity within the family (Hoot et al. 1999 Savolainen et al. 2000a, 2000b Ingrouille et al. 2002 cf. Hilu et al. 2003 see ������Discussion������). The relationship of Dilleniaceae to other angiosperms remains poorly known, despite the long-standing view that they are a line- age important for understanding angiosperm phylogeny at deep hierarchical levels. Molecular phylogenetic studies ad- dressing the relationship of Dilleniaceae within angiosperms have yet to reach a consensus regarding placement of the fam- ily beyond indicating it as an early-diverging lineage within the core group of eudicots (Stevens 2001���2009 APG 2003). A majority of these studies, however, resolve the family as sis- ter to either Caryophyllales (Hoot et al. 1999 Soltis et al. 2000, 2003, 2005, 2007) or Vitaceae (Chase et al. 1993 Savo- lainen et al. 2000b Hilu et al. 2003). Recent studies employ- ing increasingly large data sets suggest that Dilleniaceae are a yet more deeply divergent lineage within core eudicots, poten- tially sister to a superclade of multiple ordinal-level clades (of various circumscriptions, depending on the study Worberg et al. 2007 Moore et al. 2008). 1 Current address: Department of Botany and Laboratories of Analytical Biology, Smithsonian Institution, P.O. Box 37012, NMNH MRC-0166, Washington, DC 20013-7012, U.S.A. e-mail: hornj@si.edu. Manuscript received July 2008 revised manuscript received March 2009. 794 Int. J. Plant Sci. 170(6):794���813. 2009. �� 2009 by The University of Chicago. All rights reserved. 1058-5893/2009/17006-0008$15.00 DOI: 10.1086/599239

The inherent biological significance of Dilleniaceae, however, centers on its extensive structural diversity. Dilleniaceae are re- markable among angiosperm families because they vary with respect to many characters that formed the basis of the evolu- tionary trends that formerly objectified angiosperm systematics (Stebbins 1974 Stebbins and Hoogland 1976 Cronquist 1988 Takhtajan 1991). Illustrative of this, merosity within the an- droecium and gynoecium of dilleniaceous flowers ranges from 1 to 900 stamens and from 1 to 20 carpels, respectively (Gilg and Werdermann 1925 Hoogland 1952 Endress 1997). Car- pels may be entirely free or show substantial synorganization. Floral symmetry varies from polysymmetric to monosymmetric. The unusual patterns of variation among the large constellation of ������fundamental������ trend-based characters within Dilleniaceae not only shaped the former view that the family provided a link between magnoliid dicots and more ������advanced������ dicot lineages (Cronquist 1981) but also were critical for establishing evolu- tionary hypotheses within the family. Dickison (1967a, 1968, 1970b) and others (Stebbins 1974 Stebbins and Hoogland 1976 Rury and Dickison 1977) made explicit use of evolution- ary trends in their phylogenetic interpretations of structural data. Despite the family���s protean nature, the infrafamilial classifica- tion of Dilleniaceae has been rather stable, with just two major groups most often recognized. Tetracera, the only pantropical genus, along with the Neotropical endemics Curatella, Davilla, Doliocarpus, and Pinzona, constitute the tribe Delimeae (de Can- dolle 1824 ��Tetracereae: Gilg & Werdermann 1925 Hutchin- son 1964), or subfamily Delimoideae (��������Tetraceroideae������ of Hoogland 1952 and Dickison 1970b). Except for Curatella, spe- cies of these genera are often lianas, a growth form otherwise rare within the family (Kubitzki 1970, 1971). Tribe Dilleniae (de Candolle 1817, 1824), or subfamily Dillenioideae (Hoogland 1952), contains Acrotrema, Adrastaea, Didesmandra, Dillenia, Hibbertia, Pachynema, and Schumacheria, which are restricted to the Old World. The largely Australian genera Hibbertia (which often includes the monotypic Adrastaea) and Pachynema and the nearly herbaceous Acrotrema are segregated as tribes in Fig. 1 Distribution of Dilleniaceae. A, Distribution of Tetracera (Delimoideae across whole shaded area) and Doliocarpoideae (Neotropical only). B, Distribution of Hibbertia (Hibbertioideae). C, Distribution of Dillenioideae. Dillenia occurs within the whole of the shaded area Didesmandra, endemic to Borneo, is indicated by a white star distributions of other genera are as indicated. 795 HORN���PHYLOGENETICS OF DILLENIACEAE

some classifications (Gilg and Werdermann 1925 Hutchinson 1964). Dillenioideae, as considered here, display a significantly greater amount of structural and ecological diversity than Delim- oideae, and they range in habit from large rainforest trees with leaves more than 1 m long (Dillenia spp. Hoogland 1952) to xe- romorphic subshrubs possessing minute leaves 1 mm long borne on phyllocladous shoot systems (Pachynema spp. Craven and Dunlop 1992). But because the states of most characters consid- ered important for understanding relationships within Dillenia- ceae have complex distributions suggestive of mosaic evolution, uncertainty regarding the potential phylogenetic and classifica- tory significance of all structural and chemical features within the family eventually leads to the complete abandonment of any system of classification above the genus level (Gurni and Kubitzki 1981 Aymard C 1997 but see Horn 2007). Here I present a phylogenetic hypothesis of Dilleniaceae based on sequence data from four plastid loci. Of the four re- gions chosen, three have been shown to be sufficiently variable to elucidate infrafamilial relationships (rbcL: Prince and Parks 2001 Chase et al. 2002 Gustafsson et al. 2002 Wurdack et al. 2004 rps4: Reeves et al. 2001 the rpl16 intron: Kelch- ner and Clark 1997 Zhang 2000 Clausing and Renner 2001 Downie et al. 2001). Data from the plastid gene infA (Millen et al. 2001) are here employed for the first time in a phyloge- netic study. The first goal of this study is to test the monophyly of genera of Dilleniaceae and to discover the major clades within the family. The second goal is to examine character evo- lution in the context of a new phylogenetic hypothesis for the family, emphasizing characters that were important to previous hypotheses of phylogenetic trends in Dilleniaceae and angio- sperms as a whole. Finally, a new classification of Dilleniaceae, based on the phylogenetic results, is presented. Material and Methods Taxon Sampling Voucher information for the taxa included in the analyses are listed in table A1 in the online edition of the International Journal of Plant Sciences. Within Dilleniaceae, 57 species were sampled, representing 12 of the 14 genera (including major seg- regate genera) recognized in the primary specialist treatments of the family (Gilg and Werdermann 1925 Harden and Everett 1990 Aymard C 1997). Suitable material was not available for Didesmandra and Neodillenia. Twenty-two species of Hibbertia s.l., the largest and most taxonomically complex genus of the family, were included. Sampling within Hibbertia was guided by the classification of Gilg and Werdermann (1925) and by pre- liminary results of a separate molecular phylogenetic study of Hibbertia, which included a large majority of the species cur- rently recognized within the genus (Horn 2005). Within the species-rich genera Davilla, Dillenia, Doliocarpus, and Tetra- cera, a minimum of five species each were included, chosen to represent subgeneric groups recognized in recent monographs (Davilla and Doliocarpus: Kubitzki 1971 Dillenia: Hoogland 1952 Tetracera: Kubitzki 1970) and to span the geographic range. Because many recent large-scale molecular phylogenetic anal- yses suggest that Dilleniaceae are sister to either Caryophyllales or Vitaceae (see ������Introduction������), exemplars from these taxa were chosen as outgroups (table A1). Within Caryophyllales, species were chosen as exemplars of major clades (see Cuenoud �� et al. 2002). DNA Extraction, PCR Amplification, and Sequencing Leaf material for DNA extraction was obtained from sam- ples collected in the field and desiccated in silica gel and from herbarium specimens at MO, NCU, NY, and US. An area of dry leaf tissue measuring 0.5���1.0 cm2 for each sample was dis- rupted with a stainless-steel ball bearing or a porcelain bead in a Model 2000 Geno/Grinder (SPEX CertiPrep, Metuchen, NJ). Extractions of total genomic DNA were made by using the DNeasy Plant Mini Kit (Qiagen, Valencia, CA) according to the manufacturer���s protocols, except for the modification of lysis conditions by the addition of 570 mg of proteinase K (PCR grade Roche Diagnostics, Indianapolis, IN) with subse- quent agitation at 37��C for 24���48 h. Double-stranded DNA templates were amplified via PCR for four chloroplast loci: infA, rbcL, rps4, and the rpl16 intron. Amplifications and/or cycle sequencing reactions used the following primer combinations for PCR: infA, rpl36-f/rps8-r (rpl36-f: 59-CACAAATTTTACGAACGGAAG-39 rps8-r: 59-TAATGACAGAYCGAGARGCTCGAC-39 K. J. Wurdack, personal communication) rbcL, 1F/1460R (Fay et al. 1998) or rbcLDillR rps4, TRNAS/rps5 (Nadot et al. 1995) or rps4DillF/ rps4DillR rpl16 intron, F71/R1661 (Kelchner and Clark 1997). The infA primers are rooted in flanking plastid genes (i.e., rpl36 and rps8, respectively), which ensures that plastid copies are amplified, rather than nuclear copies originating from plastid-nuclear gene transfers (see Millen et al. 2001). Since the 1460R primer, typically used to amplify rbcL, primed poorly to its target sequence for a few ingroup taxa, a new reverse primer, rbcLDillR (59-TCCATACTTCACAAGCAGGAG-39), was designed for those problematic taxa. Likewise, new primers (rps4DillF: 59-GTGATCTYAGAAACCAATCKCG-39 rps4DillR: 59-ATTATTCCAACAGCAGGGCCT-39) were designed for the rps4 gene because the standard primers consistently amplified multiple fragments of varying length under a variety of PCR con- ditions for many taxa of Dilleniaceae. The newly designed rps4 primers consistently yielded single bands. Amplified products were cleaned with the QIAquick PCR Pu- rification Kit (Qiagen) and directly cycle sequenced by using an ABI BigDye Terminator Cycle Sequencing Kit (Applied Biosys- tems, Foster City, CA). Sequences were resolved on an ABI Prism 3700 DNA Analyzer automated sequencer (96 capillaries). Sequence Manipulation and Alignment Sequence assembly and editing were done with Sequencher, version 4.1.2 (Gene Codes, Ann Arbor, MI). Data matrices were aligned by eye in Se-Al, version 2.0a11 (Rambaut 1996). Within the rps4 gene, only the nucleotides within the region amplified by the rps4DillF/rps4DillR primers (excluding the priming regions themselves) were included in subsequent analy- ses. Gapped regions and regions of ambiguous alignment were identified and excluded from subsequent analyses. Gaps were interpreted as missing data and were not coded as separate characters. 796 INTERNATIONAL JOURNAL OF PLANT SCIENCES