HOME HELP FEEDBACK SUBSCRIPTIONS ARCHIVE SEARCH TABLE OF CONTENTS		QUICK SEARCH:   [advanced] Author: Keyword(s): Year:  Vol:  Page:

This Article Abstract Full Text (PDF) Submit a response Alert me when this article is cited Alert me when eLetters are posted Alert me if a correction is posted Services Email this article to a friend Similar articles in this journal Similar articles in ISI Web of Science Alert me to new issues of the journal Download to citation manager Citing Articles Citing Articles via HighWire Citing Articles via ISI Web of Science (34) Citing Articles via Google Scholar Google Scholar Articles by Judd, W. S. Articles by Olmstead, R. G. Search for Related Content PubMed Articles by Judd, W. S. Articles by Olmstead, R. G. Agricola Articles by Judd, W. S. Articles by Olmstead, R. G.

A survey of tricolpate (eudicot) phylogenetic relationships¹

²Department of Botany, University of Florida, Gainesville, Florida 32611 USA; ³Department of Biology, University of Washington, Seattle, Washington 98195 USA

Received for publication January 27, 2004. Accepted for publication June 4, 2004.

	ABSTRACT

TOP ABSTRACT INTRODUCTION BASAL TRICOLPATE RELATIONSHIPS CORE TRICOLPATES ROSIDS ASTERIDS UNRESOLVED ISSUES AND FUTURE... LITERATURE CITED

 
The phylogenetic structure of the tricolpate clade (or eudicots) is presented through a survey of their major subclades, each of which is briefly characterized. The tricolpate clade was first recognized in 1989 and has received extensive phylogenetic study. Its major subclades, recognized at ordinal and familial ranks, are now apparent. Ordinal and many other suprafamilial clades are briefly diagnosed, i.e., the putative phenotypic synapomorphies for each major clade of tricolpates are listed, and the support for the monophyly of each clade is assessed, mainly through citation of the pertinent molecular phylogenetic literature. The classification of the Angiosperm Phylogeny Group (APG II) expresses the current state of our knowledge of phylogenetic relationships among tricolpates, and many of the major tricolpate clades can be diagnosed morphologically.

Key Words: angiosperms • eudicots • tricolpates

	INTRODUCTION

TOP ABSTRACT INTRODUCTION BASAL TRICOLPATE RELATIONSHIPS CORE TRICOLPATES ROSIDS ASTERIDS UNRESOLVED ISSUES AND FUTURE... LITERATURE CITED

 
Angiosperms traditionally have been divided into two primary groups based on the presence of a single cotyledon (monocotyledons, monocots) or two cotyledons (dicotyledons, dicots). A series of additional diagnostic traits made this division useful and has accounted for the long recognition of these groups in flowering plant classifications. However phylogenetic analyses based on nuclear, plastid, and mitochondrial DNA sequences and morphology do not support this dichotomy (Donoghue and Doyle, 1989 ; Olmstead et al., 1992a ; Chase et al., 1993 ; Doyle et al., 1994 ; Doyle, 1996 , 1998 ; Mathews and Donoghue, 1999 ; Graham and Olmstead, 2000 ; Savolainen et al., 2000a ; Soltis et al., 2000 ; Hilu et al., 2003 ; Zanis et al., 2003 ). In virtually all published cladistic analyses, the "dicots" form a paraphyletic complex and their diagnostic features are mainly plesiomorphic within angiosperms (see also Soltis and Soltis, 2004 ), although the monocots do constitute a clade (Chase, 2004 ).

Nonetheless, a large number of species previously considered "dicots" do constitute a well-supported clade: the tricolpates (Donoghue and Doyle, 1989 ) or eudicots (Doyle and Hotton, 1991 ). A synapomorphy of the tricolpate clade is pollen with three apertures (tricolpate/tricolporate pollen and derivatives thereof). The tricolpate clade is the largest group of angiosperms, containing perhaps 165 000 species in just over 300 families (ca. 64% of angiosperm diversity) and encompassing phenomenal variation in morphological, anatomical, and biochemical features. The clade also is characterized by cyclic flowers and the presence of differentiated outer and inner perianth members (i.e., a calyx and corolla) may be an additional, albeit homoplasious synapomorphy (Zanis et al., 2003 ). The staminal filaments are usually slender, bearing well-differentiated anthers, and most members have S-type plastids in their sieve elements. This clade was first recognized in the morphology-based phylogenetic analysis of Donoghue and Doyle (1989) . Their monophyly was soon thereafter supported by numerous molecular analyses (Olmstead et al., 1992a ; Chase et al., 1993 ; Doyle et al., 1994 ; Soltis et al., 1997 , 2000 , 2003 ; Källersjö et al., 1998 ; Nandi et al., 1998 ; Hoot et al., 1999 ; Savolainen et al., 2000a , b ; Hilu et al., 2003 ; Zanis et al., 2003 ; Kim et al., 2004 ). This clade was first called the tricolpates (Donoghue and Doyle, 1989 ), but the name eudicots (Doyle and Hotton, 1991 ) has gained wider usage. We prefer tricolpates because this name is both descriptive and avoids a connection with the nonmonophyletic assemblage "Dicotyledoneae." Embryos having two (or more) cotyledons are not synapomorphic for this clade because they are also characteristic of Coniferales, Cycadales, Gnetales, and the basal grade of flowering plants from which monocots (which, as their name implies, share the synapomorphy of a single cotyledon) and tricolpates are derived.

During the past 15 years, our understanding of phylogenetic relationships within tricolpates has improved dramatically. This has been accomplished largely on the basis of phylogenetic analysis of molecular data, with many studies representing collaborations of several authors and combining several data sets (Chase et al., 1993 ; Chase and Cox, 1998 ; Soltis et al., 1998 ; Hilu et al., 2003 ). Thus, we now have an accurate outline (though incomplete in many details) of phylogenetic relationships within tricolpates (Fig. 1). The classifications of the Angiosperm Phylogeny Group (APG, 1998 ; APG II, 2003 ) have been based on these ongoing molecular analyses, leading to the recognition of a series of putatively monophyletic orders and families. Various secondary criteria, such as strength of support for monophyly, ease of recognition on the basis of phenotypic features, minimization of taxonomic redundancy, etc., also are used in ranking decisions (APG, 1998 ; Backlund and Bremer, 1998 ; Judd et al., 2002 ; APG II, 2003 ). The result is a classification that is phylogenetically accurate, to the extent we can presently determine, and that system is used with some slight modifications as the basis for the discussion of tricolpate diversity in this paper.

View larger version (29K): [in this window] [in a new window]   Fig. 1. Phylogenetic relationships of major groups of Tricolpates (Eudicots); modified from APG II (2003) . The names lamiids (for euasterids I) and campanulids (for euasterids II) were suggested by Bremer et al. (2002) . The names fabids (for eurosids I) and malvids (for eurosids II) are proposed here

The overall evolutionary pattern seen in tricolpates is one of hierarchically nested clades, with those recognized as families or orders in the APG systems (1998 , 2003 ) representing the "long branch" portions of the molecular-based phylogenetic trees, i.e., the well-supported clades that have been comparatively easy to "find" in DNA studies (Chase et al., 2000 ; Soltis et al., 2000 ). Most families, as circumscribed in APG II (2003) , are also easily and accurately diagnosable on the basis of their morphological characters, and have a long history of taxonomic recognition. (There are some notable exceptions, e.g., Salicaceae s. l. and several families of Lamiales, but by and large, the long-recognized families have been shown to be monophyletic, especially when obviously paraphyletic groups, such as the traditional Apocynaceae, Capparaceae, Caprifoliaceae, Ericaceae, Sapindaceae, and Tiliaceae/ Sterculiaceae/Bombacaceae, are restructured.) It has been suggested that these familial clades are the result of the canalization of suites of traits within lineages over long periods of time (Chase et al., 2000 ; Soltis et al., 2000 ). The same pattern is evident among the clades recognized at ordinal rank, although fewer of these are morphologically diagnosable. Thus many of the ordinal groupings outlined here are of novel circumscription. Relationships among familial clades within these orders are frequently problematic because the branches are short and cannot be "found" even with several genes (see for example, Ericales, Malpighiales, Saxifragales, Lamiales, and others discussed later), indicating that diversification of major lineages within these ordinal clades occurred relatively rapidly. The ordinal synapomorphies likely have been obscured by subsequent radiations, which produced the clades now recognized as families, as a result of homoplasious morphological changes within a developmentally constrained pattern of morphological variation (Soltis et al., 2000 ).

The repeated pattern of longer branches to some clades, which, in turn, possess very short branches indicates that the history of tricolpates is characterized by periodic bursts of evolution, followed by a mosaic of lineage persistence and extinction, with subsequent additional periods of rapid evolutionary diversification (Chase et al., 2000 ; Davies et al., 2004 ). We think it likely that familial and ordinal clades showing numerous morphological synapomorphies have suffered more extinction of early divergent lineages than have those clades with fewer diagnostic features (see also discussion in Chase et al., 2000 ). It is unfortunate that we know so little regarding the potential phenotypic synapomorphies for many higher-level clades, which necessarily limits our ability to understand the reasons for these episodic evolutionary diversifications.

Currently, the major supraordinal clades within tricolpates (eudicots) for which we have relatively strong support are the core tricolpates, rosids (and within rosids, the fabids and malvids), and asterids (and within asterids, the lamiids and campanulids) (Fig. 1). The phylogenetic structure of these clades, as well as their ordinal subclades, is outlined here.

	BASAL TRICOLPATE RELATIONSHIPS

TOP ABSTRACT INTRODUCTION BASAL TRICOLPATE RELATIONSHIPS CORE TRICOLPATES ROSIDS ASTERIDS UNRESOLVED ISSUES AND FUTURE... LITERATURE CITED

 
A consensus on what constitute the major groups of tricolpates is emerging from among the numerous molecular phylogenetic studies of angiosperms cited earlier (Fig. 1). Apart from a few mostly minor clades, the vast majority of species form a large clade informally designated the "core tricolpates" (or "core eudicots"). Floral form and merosity in early-diverging tricolpate lineages is highly labile, whereas floral development in core tricolpates is more strongly canalized, and except for Gunnerales, these plants usually have pentamerous flowers with two differentiated perianth whorls (Endress, 1987 ; Drinnan et al., 1994 ; Soltis et al., 2003 ; Zanis et al., 2003 ).

Branching order at the base of the core tricolpates is poorly supported (Savolainen et al., 2000b ; Soltis et al., 2000 , 2003 ; Hilu et al., 2003 ; Kim et al., 2004 ), but three large groups emerge along with several smaller lineages. The three large groups, caryophyllids, rosids, and asterids, correspond to a substantial degree with three of the major groups recognized at the rank of subclass in some late-20th century classifications (e.g., Cronquist 1981 , 1988 ; Takhtajan, 1997 ), Caryophyllidae, Rosidae, and Asteridae. However, it should be noted that there are significant differences in circumscription between the informally named monophyletic groups recognized today and those subclasses as previously delimited (see also Soltis and Soltis, 2004 ). In addition, the two remaining dicot subclasses in the Cronquist/Takhtajan systems, Hamamelidae and Dilleniidae, leave only fractured remnants in the tricolpate phylogenetic tree, distributed among the early-branching lineages at the base of the tricolpates as well as the three large clades in the core tricolpates.

Ranunculales are well supported as sister to the remaining tricolpates, with Proteales and Sabiaceae the subsequent sisters to other tricolpates, although the branching order is unclear, and in some analyses Proteales and Sabiaceae instead form a clade sister to other tricolpates (Nandi et al., 1998 ; Hoot et al., 1999 ; Doyle and Endress, 2000 ; Savolainen et al., 2000b ; Soltis et al., 2000 , 2003 ; Hilu et al., 2003 ; Zanis et al., 2003 ; Kim et al., 2004 ). Monophyly of Ranunculales is supported by DNA sequence studies (Chase et al., 1993 ; Drinnan et al., 1994 ; Hoot and Crane, 1995 ; Soltis et al., 1997 , 1998 , 2000 ; Källersjö et al., 1998 ; Hoot et al., 1999 ; Savolainen et al., 2000a , b ; Hilu et al., 2003 ) and morphology (Loconte et al., 1995 ), and synapomorphies may include presence of the alkaloid berberine and reduced fiber-pit borders. Ranunculales previously have been associated with woody magnoliids (e.g., Cronquist, 1981 ) because their flowers have free parts that are sometimes spirally arranged. Cronquist (1981) suggested that the connection to the woody magnoliids was via Illiciaceae (ANITA grade; Qiu et al., 1999 ) due to the presence of triaperturate pollen in both groups. Illiciaceae and the related Schisandraceae are now recognized as the sole exception to the exclusive presence of triaperturate pollen in the tricolpates and are not closely related to them.

Some analyses have recovered Papaveraceae (including Fumariaceae) sister to the remaining families of Ranunculales (Hoot and Crane, 1995 ; Hoot et al., 1999 ; Soltis et al., 2000 ), whereas others (Hilu et al., 2003 ; Kim et al., 2004 ) support Eupteleaceae (trees with reduced, wind-pollinated flowers) as sister to the rest, with Papaveraceae the next family to diverge. Ranunculaceae apparently are sister to Berberidaceae, although no morphological synapomorphy is evident (Hoot and Crane, 1995 ; Kim and Jansen, 1995 ; Hoot et al., 1999 ; Kim et al., 2004 ). In contrast to most Ranunculaceae, Berberidaceae have only a single carpel. Menispermaceae, a group of vines or lianas with imperfect flowers and drupaceous fruits, are probably sister to the Berberidaceae + Ranunculaceae clade. Lardizabalaceae, which apparently evolved the climbing habit independently, are sister to that more inclusive clade.

Proteales, with Nelumbonaceae sister to Proteaceae + Platanaceae, represent a novel grouping that had not been associated prior to the acquisition of molecular data (Olmstead et al., 1992a ; Chase et al., 1993 ; Hoot et al., 1999 ; Soltis et al., 2000 ; Hilu et al., 2003 ), but gynoecia with one or two pendent ovules per carpel are putatively synapomorphic for these plants. Synonymy of Platanaceae under Proteaceae is recommended in APG II, because extant Platanaceae are monogeneric and, thus, their inclusion in an expanded Proteaceae would reduce taxonomic redundancy. However, other genera of Platanaceae are known as fossils and even the extant species of Platanus belong to two morphologically distinct subgenera. Thus, the characterization of Platanaceae as "monogeneric" is misleading about their past diversity. In addition, an expanded Proteaceae would be difficult to diagnose morphologically.

Proteales, Sabiaceae, Buxaceae, Didymelaceae, and Trochodendraceae (Trochodendron and Tetracentron) form a grade (Soltis et al., 2000 , 2003 ; Hilu et al., 2003 ; Kim et al., 2004 ) sister to the core tricolpates. Trochodendraceae are united by their distinctive flattened seeds and laterally connate carpels that are nectiferous and protrude abaxially forming follicle-like fruits. Buxaceae and Didymelaceae (now both optionally included in Buxaceae in APG II, but treated here together as Buxales) are united by encyclocytic stomata and racemose inflorescences of small, imperfect flowers with a biseriate perianth of tepals. Their stigmas are decurrent, extending the entire length of the style. Recent evidence indicates that Buxaceae + Didymelaceae may be sister to core tricolpates (Hilu et al., 2003 ).

	CORE TRICOLPATES

TOP ABSTRACT INTRODUCTION BASAL TRICOLPATE RELATIONSHIPS CORE TRICOLPATES ROSIDS ASTERIDS UNRESOLVED ISSUES AND FUTURE... LITERATURE CITED

 
Gunnerales are well supported as sister to all other core tricolpates (Hilu et al., 2003 ; Soltis et al., 2003 ). Gunnerales are dioecious, have reduced, two-merous flowers, and have either bisporic or tetrasporic megagametophyte development. In contrast, the remaining core tricolpates, consisting of Berberidopsidales, Santalales, Dilleniaceae, caryophyllids, Saxifragales, rosids, and asterids, usually have pentamerous (or less commonly, tetramerous) flowers, and predominantly show the Polygonum-type megagametophyte development.

It is noteworthy that phylogenetic analyses of angiosperm MADS-box genes show two gene clades within the core tricolpates, i.e., euAP1 (including Arabidopsis APETALA1 and Antirrhinum SQUAMOSA) and euFUL (including Arabidopsis FRUITFULL), whereas non-core tricolpate clades have only sequences similar to euFUL genes (Litt and Irish, 2003 ). This indicates that a duplication event occurred in these floral genes in the common ancestor of the core tricolpates. The euAP1 gene clade includes key regulators of floral development, which have been implicated in the specification of perianth identity. The presence of euAP1 genes only in core tricolpates indicates that there may have been changes in developmental mechanisms that are correlated with the fixation of floral structures seen in this clade (Litt and Irish, 2003 ). In addition, the core tricolpates have evolved a divergent APETALATA3 C-terminal domain, which is correlated with the acquisition of a role in specifying perianth structures (Lamb and Irish, 2003 ).

Relationships among the core tricolpate clades are still unclear (Hoot et al., 1999 ; Savolainen et al., 2000b ; Soltis et al., 2000 , 2003 ; Hilu et al., 2003 ; Kim et al., 2004 ), as illustrated in the uncertainties regarding the phylogenetic position of the small Southern Hemisphere clade, Berberidopsidales.

Berberidopsidales (Fig. 1), comprising Aextoxicaceae and Berberidopsidaceae, are a well-supported clade (Soltis et al., 2000 , 2003 ; Kim et al., 2004 ); they were not recognized by APG II (2003) because their relationships are still unclear. These two families are distinctive: Aextoxicaceae are dioecious and have opposite leaves that, along with the ovary, are covered with peltate scales; Berberidopsidaceae have bisexual flowers, a scrambling habit with alternate, often spinose-serrate leaves, and flowers with parietal placentation. They share the possible apomorphies of stout filaments and presence of a ring of vascular bundles in the petiole. The order may be sister to the rosids plus Saxifragales and Vitales (Soltis et al., 2000 ), caryophyllids (Savolainen et al., 2000b ), asterids plus Santalales (Soltis et al., 2003 ; Kim et al., 2004 ), asterids plus caryophyllids and Santalales (Hilu et al., 2003 ), or just asterids (Kim et al., 2004 ).

Dilleniaceae, another isolated clade, may be the sister group of the caryophyllids (Soltis et al., 2000 , 2003 ; APG II, 2003 ) or Vitales (Hilu et al., 2003 ). They are distinctive in that their often showy, pentamerous flowers have the petals crumpled in the bud, numerous stamens (a secondary increase), and separate carpels. Like many caryophyllids, the sepals are persistent, and often accrescent. Santalales are another clade of uncertain position. Placements indicated by molecular studies include sister to caryophyllids (Soltis et al., 2000 ) or caryophyllids plus asterids (Hilu et al., 2003 ). Monophyly of Santalales is based on molecular studies (Nickrent and Soltis, 1995 ; Källersjö et al., 1998 ; Savolainen et al., 2000b ; Soltis et al., 2000 , 2003 ; Hilu et al., 2003 ) and the presence of polyacetylenes, roots lacking root hairs, one-seeded, indehiscent fruits, and seeds with the coat reduced/crushed. The parasitic habit (with conventional roots replaced by developmentally complex haustoria, which connect the parasite to the host) is present in most, and it certainly evolved early in the history of the clade, but is nonetheless derived. Santalales also lack mycorrhizal associations, and the stamens are typically opposite the petals. Delimitations of families within this clade are problematic, with both "Olacaceae" and "Santalaceae," as previously circumscribed, being paraphyletic: the rest of the order is embedded in "Olacaceae," whereas Viscaceae are part of the "Santalaceae." For this reason, Viscaceae, despite their phenetic distinctiveness (e.g., twig epiphytism), are sometimes placed within an expanded Santalaceae (APG II, 2003 ; but compare with Nickrent, 2003 ).

Caryophyllids are treated as Caryophyllales by APG II (2003) , whereas this clade is divided into two orders, i.e., Caryophyllales and Polygonales, in Judd et al. (2002) . The monophyly of caryophyllids is supported by molecular data (Soltis et al., 2000 , 2003 ; Hilu et al., 2003 ) and possibly by anther wall development and simple, nonbordered perforation plates (but the latter may be a synapomorphy of a more inclusive group including Santalales). There are also similarities in the type of sieve-element plastids (Behnke, 1999 ). The lack of clear morphological support for this clade leads us to prefer its division into two orders: Caryophyllales sensu stricto and Polygonales (see also Hilu et al., 2003 ).

Caryophyllales s. s. largely correspond to previous circumscriptions of this order (Cronquist, 1981 ; Takhtajan, 1997 ; Thorne, 2001 ), with the small families Simmondsiaceae, and Asteropeiaceae plus Physenaceae, as successive sister groups of the core members. The monophyly of Caryophyllales s. s. is moderately to strongly supported on the basis of DNA sequences (Soltis et al., 2000 , 2003 ; Cuénoud et al., 2002 ; Hilu et al., 2003 ) as well as morphology. Successive cambia and unilacunar nodes are synapomorphic for the order, with core Caryophyllales (i.e., all families except Simmondsiaceae, Asteropeiaceae, and Physenaceae) diagnosed by numerous additional derived characters, including P-type plastids with a ring of proteinaceous filaments and a central angular crystal, betalains, a single whorl of tepals, free-central/basal placentation, pollen with spinulose and tubuliferous/punctate ektexine, and seeds with perisperm and curved embryo (Mabry, 1973 ; Behnke, 1976 , 1994 ; Eckardt, 1976 ; Cronquist, 1981 ; Rodman et al., 1984 ; Rodman, 1990 , 1994 ; Bittrich, 1993 ). These families also typically have separate styles or well-developed style branches. Within core Caryophyllales, phylogenetic relationships and family circumscriptions are not clear and circumscriptions of some commonly recognized families have been altered (Judd et al., 2002 ; APG II, 2003 ; Stevens, 2003 ). Amaranthaceae now include Chenopodiaceae; both share the loss of the central protein crystal in their sieve element plastids. Phytolaccaceae, as traditionally circumscribed, may not be monophyletic, but are likely closely related to Sarcobataceae and Nyctaginaceae, a group characterized by raphides. Portulacaceae (Judd et al., 2002 ) are blatantly paraphyletic, having given rise to Cactaceae, Didiereaceae, and possibly Basellaceae. Portulacaceae, Cactaceae, and Didiereaceae are nearly all succulents with crassulacean acid metabolism (CAM) metabolism. The monophyly of Cactaceae is supported by numerous characters, including their conspicuous differentiated shoots, with the leaves of the short shoots forming spines (Judd et al., 2002 ). Betalains apparently have been lost (and anthocyanins re-acquired) more than once in the order, because Caryophyllaceae and Molluginaceae, which have anthocyanins, are not closely related (Soltis et al., 2000 ).

Polygonales are here defined broadly, including Plumbaginaceae, Polygonaceae, Droseraceae, Nepenthaceae, and relatives. The monophyly of the order is supported by the putative apomorphies of plumbagin (a naphthoquinone) and related compounds, vascularized glandular hairs, starchy endosperm, and possibly basal placentation (Williams et al., 1994 ; Soltis et al., 2000 , 2003 ; Hilu et al., 2003 ). These families comprise two major subclades. The first includes Polygonaceae, Plumbaginaceae, Tamaricaceae, and Frankeniaceae and is supported by sulphinated flavonols and vessels with minute lateral wall pits. The second includes Droseraceae, Nepenthaceae, Drosophyllaceae, Dioncophyllaceae, and Ancistrocladaceae and is supported by carnivorous habit, coiled leaves, and corolla contorted in bud (Meimberg et al., 2000 ; Soltis et al., 2000 , 2003 ; Cuénoud et al., 2002 ; Hilu et al., 2003 ). The carnivorous habit evidently has been lost in Ancistrocladaceae and two of the three genera of Dioncophyllaceae, to which the former is clearly sister, and the two share the lianous habit, sclerenchyma bundles in their petioles, actinocyclic stomata, and the loss of cortical vascular bundles in the stems.

The remaining species of core tricolpates belong to the rosid clade, asterid clade, or Saxifragales (Fig. 1). Saxifragales are possibly sister to the rosids (Soltis et al., 2000 ; Savolainen et al., 2000a ); however, Saxifragales were found to be sister to the caryophyllids based on the data set of Soltis et al. (2003) and Kim et al. (2004) or in a polytomy with rosids, Vitales, Dilleniaceae, Berberidopsidales, Santalales, caryophyllids, and asterids by Hilu et al. (2003) . Saxifragales are a grouping not appreciated until the advent of recent molecular phylogenetic analyses (Chase et al., 1993 ; Morgan and Soltis, 1993 ; Soltis and Soltis, 1997 ; Soltis et al., 1997 , 2000 , 2003 ; Hoot et al., 1999 ; Savolainen et al., 2000b ; Hilu et al., 2003 ; Davis and Chase, 2004 ). The group is morphologically heterogeneous, and its members were placed by Cronquist (1981) into three different subclasses. Lack of resolution within the order makes determination of morphological synapomorphies especially difficult, but the group is characterized by having the floral apex concave early in development (Soltis and Hufford, 2002 ; Fishbein et al., in press ) and carpels that are free, at least apically (Stevens, 2003 ). Paeoniaceae, Peridiscaceae, Hamamelidaceae, Altingiaceae, and Cercidophyllaceae are isolated and of unclear relationships to the other families of the order. Saxifragaceae, themselves, may be most closely related to Grossulariaceae, Iteaceae, and Pterostemonaceae (Fishbein et al., 2001 ), all of which have been included in Saxifragaceae s. l. in the past. These plants all have flowers with variously developed hypanthia, although a hypanthium may have evolved independently in a few other members of the order. A second clear clade within this order includes Crassulaceae (succulents) and the small families Haloragaceae, Penthoraceae, Tetracarpaeaceae, and Aphanopetalaceae (Fishbein et al., 2001 ). Members of this clade lack stipules and have unilacunar nodes. Penthoraceae, Tetracarpaeaceae, and Aphanopetalaceae are each monogeneric, and in APG II (2003) are optionally included in Haloragaceae, but such a broadly defined Haloragaceae are not now known to be morphologically diagnosable.

	ROSIDS

TOP ABSTRACT INTRODUCTION BASAL TRICOLPATE RELATIONSHIPS CORE TRICOLPATES ROSIDS ASTERIDS UNRESOLVED ISSUES AND FUTURE... LITERATURE CITED

 
Rosids comprise a heterogeneous grouping of orders supported by DNA-based phylogenetic analyses (Soltis et al., 2000 , 2003 ; Hilu et al., 2003 ). Most members of this group belong to one of two major subclades (Fig. 1), which are called fabids (Zygophyllales, Celastrales, Malpighiales, Oxalidales, Fabales, Rosales, Cucurbitales, and Fagales) and malvids (Brassicales, Malvales, and Sapindales; Fig. 1). These names were first suggested by J. Reveal at too late a date to be included in APG II to replace eurosids I and eurosids II, respectively. The position of Vitaceae (including Leea, sometimes placed in a separate family, Leeaceae) is unclear. The family is possibly sister to the rest of the rosids (Soltis et al., 2000 ; Kim et al., 2004 ), but may belong elsewhere nearby in the poorly resolved nexus at the base of the core tricolpates (e.g., Soltis et al., 2003 ; Hilu et al., 2003 ). Thus placement of Vitaceae in the rosids in APG II (2003) may not hold up. Other groups belonging to rosids, but not placed in the two large clades fabids and malvids, include several small families (Aphloiaceae, Geissolomataceae, Ixerbaceae, Picramniaceae, Strasburgeriaceae) and Crossosomatales, Geraniales, and Myrtales.

Crossosomatales, as defined in APG II (2003) , contain Crossosomataceae, Stachyuraceae, and Staphyleaceae, and are characterized by a distinctive seed coat, in which the cell walls of the many-layered testa are more or less all lignified. The order possibly could be expanded to include Geossolomataceae, Ixerbaceae, and Strasburgeriaceae, which have a similar testa anatomy (APG II, 2003 ; Stevens, 2003 ). Geraniales include Geraniaceae and a few small families (APG II, 2003 ) having leaf margins with glandular teeth, vessels with simple perforations, obdiplostemonous flowers with a persistent calyx, and nectaries positioned on the outside of the androecium (Stevens, 2003 ). All of the families of Geraniales sensu Cronquist (1981) , except Geraniaceae, are placed elsewhere in APG II (2003) .

Myrtales were placed in malvids (eurosids II) in APG (1998) , but they were left unplaced within the rosids in APG II (2003) . Myrtales, along with Geraniales, are weakly associated with malvids in some studies (e.g., Savolainen et al., 2000b ; Soltis et al., 2003 ; Hilu et al., 2003 ), or with fabids in others (e.g., Soltis et al., 2000 ). Monophyly of Myrtales is indicated by their vessel elements with vestured pits, stems with internal phloem, stipules absent or present only as small lateral or axillary structures, flowers with a short to elongate hypanthium, stamens incurved in bud, and a single style (the carpels being nearly completely connate) and receives strong molecular support (Conti et al., 1996 , 1997 ; Savolainen et al., 2000b ; Soltis et al., 2000 ). Vochysiaceae are sister to Myrtaceae, a family characterized by pellucid dots (with aromatic terpenoids). Lythraceae and Onagraceae are sister, sharing a valvate calyx and grouped vessels in the wood, and Combretaceae are sister to them (Johnson and Briggs, 1984 ; Conti et al., 1996 , 1997 ; Savolainen et al., 2000b ; Soltis et al., 2000 , 2003 ). Melastomataceae are sister to Memecylaceae (Clausing and Renner, 2001 ), and both have unusual anthers: those of Melastomataceae usually opening by pores and often bearing various appendages and those of Memecylaceae with a resin-secreting gland on the connective. The small families Rhynchocalycaceae, Oliniaceae, Penaeaceae, Alzateaceae, and Crypteroniaceae constitute a clade (Conti et al., 1999 ), the members of which have square stems with more or less swollen nodes, stamens equal in number to and opposite the petals, exotestal cells periclinally elongated, and a nonfibrous endotegmen. Stevens (2003) has suggested that these could be combined.

Fabids (eurosids I)
The APG II (2003) circumscription of fabids includes seven orders (Celastrales, Cucurbitales, Fabales, Fagales, Malpighiales, Oxalidales, and Rosales) and two families unassigned to the order. The fabids were first recognized as "rosids I" by Chase et al. (1993) , but with Myrtales and part of Geraniales included, and later renamed "eurosids I" with the present circumscription (APG, 1998 ; APG II, 2003 ). One of the unassigned families is Zygophyllaceae, including the hemiparasitic Krameria. These are sister groups and may be retained as separate families, together forming Zygophyllales (Soltis et al., 2000 ). It does not seem desirable to treat Krameria within an expanded Zygophyllaceae, because the expanded family would have no clear morphological apomorphies. Huaceae are the other unassigned family; they are placed consistently near Celastrales (Savolainen et al., 2000b ; Soltis et al., 2000 ), but with weak support. Celastrales, Malpighiales, and Oxalidales may form a clade, as do Fabales, Cucurbitales, Rosales, and Fagales (Soltis et al., 1995 ; APG II, 2003 ; Hilu et al., 2003 ). It is noteworthy that the latter clade has some members of each order having nitrogen-fixing nodules on their roots (APG II, 2003 ).

Celastrales, as defined on the basis of molecular data (APG II, 2003 ), now comprise Celastraceae, Lepidobotryaceae, and Parnassiaeae (optionally including Lepuropetalon; Savolainen et al., 2000a ; Soltis et al., 2000 ). Like Zygophyllales, the group is difficult to characterize morphologically. Oxalidales, including Brunelliaceae, Cephalotaceae, Connaraceae, Cunoniaceae, Elaeocarpaceae, and Oxalidaceae, are another order that has a novel circumscription based on recent phylogenetic analyses and are difficult to diagnose. Stevens (2003) noted that these plants have epitropous ovules with a multiplicative outer integument and seeds with a fibrous/tracheidal exotegmen and a crystalliferous and palisade endotesta.

The grouping of ca. 37 families currently recognized as Malpighiales (APG II, 2003 ) was not apparent prior to DNA-based phylogenetic analyses (Chase et al., 1993 , 2002 ; Savolainen et al., 2000b ; Soltis et al., 2000 ; Hilu et al., 2003 ), although members of the clade often have toothed leaves, with the teeth having a single vein running into a congested and often deciduous apex (i.e., teeth violoid, salicoid, or theoid; see Hickey and Wolfe, 1975 ; Judd et al., 2002 ). Relationships within this large clade are poorly resolved. However, several small monophyletic groups of families can be inferred. Passifloraceae, Malesherbiaceae, Turneraceae (optionally a single family, Passifloraceae, in APG II, 2003 ), and possibly also Achariaceae, form a clade supported by the presence of cyclopentenoid cyanogenic glucosides (or cyclopentenyl fatty acids), with the first three likely to be linked by the apomorphies of a hypanthium that does not involve the androecium, and a corona, the latter especially well developed in Passifloraceae. Turneraceae and Malesherbiaceae have similar reticulate seeds. Rhizophoraceae and Erythroxylaceae share tropane and pyrrolidin alkaloids, terminal buds protected by stipules, colleters, and green embryos and were optionally considered one family, Rhizophoraceae, in APG II (2003) . Clusiaceae s. l. (Guttiferae; probably including Bonnetiaceae and Hypericaceae, although these are maintained in APG II, 2003 ) and the distinctive aquatic Podostemaceae share secretory canals and xanthone pigments. Podostemaceae may be nested within Clusiaceae s. l. (Gustafsson et al., 2002 ; Chase et al., 2002 ). Ochnaceae, Medusagynaceae, and Quiinaceae have vestured pits, tenuinucellate ovules, and lack floral nectaries; medullary bundles occur in both Ochnaceae and Medusagynaceae. These three families were optionally also proposed by APG II as the single family Ochnaceae. Malpighiaceae are sister to Elatinaceae, and both have glandular hairs on their opposite or whorled leaves and a base chromosome number of six (Davis and Chase, 2004 ). Finally, Chrysobalanaceae, Euphroniaceae, Dichapetalaceae, Trigoniaceae, and Balanopaceae form a strongly supported clade (Soltis et al., 2000 ). These plants have only two ovules per carpel and, except for Balanopaceae, which is sister to the rest, vestured pits, paracytic stomata, usually zygomorphic flowers, and tenuinucellate ovules (Stevens, 2003 ). All of these relationships receive molecular support (Litt and Chase, 1999 ; Savolainen et al., 2000a ; Soltis et al., 2000 ; Chase et al., 2002 ; Davis and Chase, 2004 ). These same analyses indicate that Flacourtiaceae and Euphorbiaceae, as previously circumscribed, are not monophyletic. The cyanogenic members of "Flacourtiaceae" are now treated as Achariaceae. A major change, especially for temperate-zone botanists, is the transfer of the non-cyanogenic taxa (i.e., those with salicoid teeth) to Salicaceae (Judd, 1997 ; Chase et al., 2002 ; Boucher et al., 2003 ), thus dramatically expanding this otherwise homogeneous family. Salicoid teeth have a single vein entering the tooth, expanding at the apex, which is associated with a spherical, glandular structure (Nandi et al., 1998 ; Judd et al., 2002 ).

The subfamilies of Euphorbiaceae s. l. with two ovules per locule have been segregated (Savolainen et al., 2000a ; Chase et al., 2002 ) as Picrodendraceae (spiny pollen), Phyllanthaceae (divided styles, schizocarps, often with dimorphic branches), and Putranjivaceae (mustard oils). This has resulted in a monophyletic and more homogeneous Euphorbiaceae s. s. (APG II, 2003 ; see also Judd et al., 2002 ; Stevens, 2003 ) having gynoecia with only one ovule per locule and, except for Acalyphoideae, laticifers with colored or milky latex. Imperfect flowers, divided styles, and trilobed, schizocarpic fruits apparently evolved independently in Euphorbiaceae and Phyllanthaceae or were lost in the putatively related families (which at present are not clear).

The core of Cucurbitales (Cucurbitaceae, Tetramelaceae, Datiscaceae, and Begoniaceae) represent a close-knit group recognized in previous classifications (e.g., Cronquist, 1981 ) and share similarities of stems with separate vascular bundles, cucurbitacins (oxidized triterpenes), palmate leaves with more or less cucurbitoid teeth, and imperfect flowers with an inferior ovary and often intruded parietal placentation (Judd et al., 2002 ; Stevens, 2003 ). In both Cucurbitaceae and Begoniaceae, the androecium and gynoecium converge in appearance, the yellow, twisted stigmas of Begoniaceae resembling the stamens and the contorted and usually connate anthers of Cucurbitaceae likely mimicking the stigma (Judd et al., 2002 ). Recent molecular studies (Savolainen et al., 2000a ; Soltis et al., 2000 ; Hilu et al., 2003 ) have reinforced the monophyly of this group and inclusion in Cucurbitales (APG II, 2003 ) of Anisophylleaceae, Coriariaceae, and Corynocarpaceae.

Quillajaceae, Surianaceae, Polygalaceae, and Fabaceae comprise Fabales, another order that was not recognized until the advent of molecular data (Chase et al., 1993 ; Savolainen et al., 2000b ; Soltis et al., 2000 ; Kajita et al., 2001 ). Morphological synapomorphies are unclear, but these plants have green embryos. Some also have fluorescing wood. Phylogenetic relationships within this order are still unclear. The zygomorphic flowers of Polygalaceae are superficially like those of Fabaceae subfamily Faboideae, but in detail they are different (Judd et al., 2002 ). Fabaceae (or Leguminosae) are monophyletic (Chappill, 1994 ; Doyle, 1994 ; Doyle et al., 1997 ; Kajita et al., 2001 ) and nitrogen fixation occurs in most species, but is lacking in many early-diverging lineages; it is homoplasious, and not synapomorphic for the family. Three subgroups are usually recognized within Fabaceae: "Caesalpinioideae," Mimosoideae, and Faboideae (= Papilionoideae). In most classifications (e.g., Pohill et al., 1981 ), these are considered subfamilies. Phylogenetic analyses of both morphological (Chappill, 1994 ; Tucker and Douglas, 1994 ) and molecular data (Doyle, 1987 ; Doyle et al., 1997 ; Bruneau et al., 2001 ; Kajita et al., 2001 ; Doyle and Luckow, 2003 ) have indicated that "Caesalpinioideae" are paraphyletic; Cercis and its immediate relatives, including Bauhinia, probably are sister to the rest of the family.

Fagales represent the core of the "Englerian" Amentiferae (Stern, 1973 ; Hamamelidae of Cronquist, 1981 ) and are monophyletic and easily recognized (Chase et al., 1993 ; Manos et al., 1993 ; Källersjö et al., 1998 ; Savolainen et al., 2000b ; Soltis et al., 2000 ; Hilu et al., 2003 ). Their flowers are unisexual, typically in spikes or catkins, have a reduced or missing perianth and inferior ovaries, and the plants have gland-headed and/or stellate hairs (Hufford, 1992 ; Judd et al., 2002 ). Nothofagus (Nothofagaceae) is distinct from Fagaceae and sister to the remaining families of Fagales, with Fagaceae sister to the rest (Manos et al., 1993 ; Manos and Steele, 1997 ; Hilu et al., 2003 ). The remaining families (Betulaceae, Casuarinaceae, Juglandaceae, Myricaceae, Rhoipteleaceae, and Ticodendraceae) are united by their triporoporate pollen and represent the extant members of the fossil Normapolles complex (Kedves, 1989 ). Juglandaceae and Rhoipteleaceae (the latter an optional synonym of the former; APG II, 2003 ) are sisters, forming a clade diagnosed by pinnately compound leaves; these two, along with Myricaceae, may form a clade on the basis of their aromatic glandular hairs and gynoecia with a single orthotropous ovule (Savolainen et al., 2000a ; Judd et al., 2002 ). Betulaceae are sister to Ticodendraceae, and both have two-ranked leaves and clusters of sclerids in the bark; this clade is then sister to Casuarinaceae (Manos and Steele, 1997 ; Savolainen et al., 2000a ; Soltis et al., 2000 ), which share pollen grains with the exine having tiny spines in rows. The cupules of Nothofagus are unlikely to be homologous with those of Fagaceae, and doubly serrate leaves and horizontal lenticels apparently have evolved independently in Nothofagaceae and Betulaceae.

Monophyly of Rosales has received strong support from analyses of DNA sequences (Chase et al., 1993 ; Källersjö et al., 1998 ; Savolainen et al., 2000b ; Soltis et al., 2000 ; Sytsma et al., 2002 ; Hilu et al., 2003 ). The circumscription of this order in APG II (2003) is divergent from that of Cronquist (1981 , 1988 ) and other evolutionary-taxonomic classifications, which used the order as something of a taxonomic dumping ground for taxa with plesiomorphic pentamerous flowers. Cronquist (1981) stated that the order was exceedingly diverse and included "what is left over after all the more advanced, specialized orders [of Rosidae] have been delimited." He included in Rosales, for example, Connaraceae, Cephalotaceae, and Cunoniaceae (here in Oxalidales), Dialypetalanthaceae (embedded in Rubiaceae of Gentianales; Fay et al., 2000 ), Pittosporaceae (Apiales), Byblidaceae sensu lato (divided and placed in Ericales and Lamiales), Hydrangeaceae (Cornales), Columelliaceae (campanulids), Neuradaceae (Malvales), Chrysobalanaceae (Malpighiales), Crossosomataceae (Crossosomatales), Grossulariaceae (as narrowly delimited, in Saxifragales, but as considered by Cronquist, with genera now placed in several other families and orders), Crassulaceae and Saxifragaceae (Saxifragales), Surianaceae (Fabales), and Rhabdodendraceae (caryopyllids)! Rosales, as here delimited, include only one family in common with Cronquist's Rosales—Rosaceae! As currently delimited (APG II, 2003 ), the order also includes Rhamnaceae, Elaeagnaceae, Dirachmaceae, and Barbeyaceae, and also the former families of Urticales (Ulmaceae, Cannabaceae, including Celtis and relatives, Moraceae, and Urticaceae, including Cecropiaceae). Rosales are still heterogeneous morphologically, but a reduction (or lack) of endosperm may be synapomorphic. The presence of a hypanthium may also be synapomorphic, and this structure is evident in Rosaceae, Rhamnaceae, Elaeagnaceae, Dirachmaceae, and some Ulmaceae. Loss of the hypanthium probably is a synapomorphy of the clade comprised of Cannabaceae, Moraceae, and Urticaceae. Rosaceae are sister to the remaining families and are plesiomorphic, but their monophyly may be supported by numerous stamens; molecular studies have supported their monophyly (Morgan et al., 1994 ; Evans et al., 2000 ). Families such as Rhamnaceae and Elaeagnaceae, although bearing showy flowers with an obvious hypanthium, are more closely related to the former Urticales than to Rosaceae. This relationship is supported by DNA sequences and a trans-spliced intron in the mitochondrial gene nad1 (Qiu et al., 1998 ), as well as the possession of single whorl of stamens opposite the petals. The monophyly of the urticoid subclade within Rosales (APG II, 2003 ) is well supported (Savolainen et al., 2000b ; Soltis et al., 2000 , 2003 ; Sytsma et al., 2002 ; Hilu et al., 2003 ), and they have long been recognized by their leaves with urticoid teeth, at least one prominent prophyllar bud, reduced, inconspicuous flowers with five or fewer stamens, and bicarpellate unilocular ovaries with a single apical (to basal) ovule (Humphries and Blackmore, 1989 ; Judd et al., 1994 , 2002 ; Stevens, 2003 ). Ulmaceae are probably sister to a clade containing Cannabaceae (including Celtidaceae), Moraceae, and Urticaceae. These three families share imperfect flowers, curved embryos, and the loss of a hypanthium. Cannabaceae (Celtidaceae, in Judd et al., 2002 ) are sister to Moraceae + Urticaceae, the members of which share laticifers and milky latex (reduced in many Urticaceae, a family which should include Cecropia and relatives; Sytsma et al., 2002 ).

Malvids (eurosids II)
The APG II (2003) circumscription of malvids includes three orders (Brassicales, Malvales, and Sapindales) and one unplaced family (Tapisciaceae). The malvids were first recognized as "rosids II" by Chase et al. (1993) , but with Crossomatales and part of Geraniales and included; this clade was later renamed "eurosids II" and included Myrtales, but not Geraniales or Crossosomatales (APG, 1998 ).

Brassicales are characterized by the presence of glucosinolates, which react with myrosinase (usually contained in specialized spherical myrosin cells) to release mustard oils (isothiocynates; Dahlgren, 1977 ; Rodman, 1981 , 1991 ; Rodman et al., 1993 , 1996 ; Judd et al., 1994 , 2002 ; Stevens, 2003 ), but indeterminate (racemose) inflorescences are also diagnostic and, at the cellular level, an endoplasmic reticulum with dilated cisternae provides additional support for the order (Jorgensen, 1981 ). It is noteworthy that glucosinolates have also evolved in Drypetes and Putranjiva (Putranjivaceae) of Malpighiales. Monophyly of Brassicales also is supported by DNA studies (Rodman et al., 1993 , 1996 ; Soltis et al., 1997 , 2000 , 2003 ; Källersjö et al., 1998 ; Karol et al., 1999 ; Savolainen et al., 2000b ; Hilu et al., 2003 ). Despite their monophyly and chemical distinctiveness, Brassicales are morphologically heterogeneous, as is evident from the fact that Cronquist (1981) placed these families in seven orders (i.e., Batales, Capparales, Celastrales, Geraniales, Polygalales, Sapindales, and Violales, in two subclasses, Dilleniidae and Rosidae). Akaniaceae and Tropaeolaceae are sisters and have large zygomorphic flowers with a hypanthium. Moringaceae and Caricaceae form another clade, which, with the previous group, form the first two groups successively sister to the rest of Brassicales. These members of the order have pentamerous flowers and seeds with endosperm and a straight embryo. Most core members of Brassicales (e.g., Bataceae, Salvadoraceae, Koeberliniaceae, Resedaceae, Tovariaceae, and Brassicaceae s. l.) have tetramerous flowers and seeds with a curved embryo that more or less lacks endosperm; they also have vestured pits in the vessel elements. Bataceae and Salvadoraceae are sisters, sharing opposite leaves with two-trace nodes and paracytic stomata; these two families, along with Koeberliniaceae, form a clade that shows adaptations to dry and/or salty habitats; they also have a base chromosome number of 11 (Stevens, 2003 ). Brassicaceae (including Capparaceae) are by far the largest family in the order, and their monophyly is strongly supported by DNA sequences (Soltis et al., 2000 ; Hall et al., 2002 ) as well as morphology (Judd et al., 1994 , 2002 ; Stevens, 2003 ). Both molecular and morphological data indicate that Capparaceae, as previously circumscribed (Cronquist, 1981 ; Kubitzki, 2003 ), are nonmonophyletic. The family comprises three clearly differentiated and monophyletic subfamilies: Capparoideae, Cleomoideae, and Brassicoideae (Hall et al., 2002 ). The question of whether or not to treat these three clades as subfamilies (as here and in APG II, 2003 ) or as closely related families (as recommended by Hall et al., 2002 ) is not easily answered because there are no unambiguous criteria for determination of rank.

Malvales are clearly monophyletic (Alverson et al., 1998 ; Fay et al., 1998a ; Källersjö et al., 1998 ; Bayer et al., 1999 ; Savolainen et al., 2000b ; Soltis et al., 2000 ; Kubutzki and Chase, 2003 ) and may be diagnosed by their stratified phloem with fibrous and soft layers, wedge-shaped rays in the wood, mucilage canals and cavities, stellate hairs, malvoid teeth, and cyclopropenoid fatty acids (Judd and Manchester, 1998 ; Stevens, 2003 ). The stamens are usually numerous (by secondary increase, with centrifugal initiation. Stevens (2003) noted that the seeds of Malvales have the exotegmen thickened, lignified, and with a palisade. Basal nodes within Malvales are unsupported in the DNA studies, but include the following lineages: (1) Bixaceae (including Cochlospermaceae and Diegodendonaceae), (2) Cistaceae, Dipterocarpaceae, and Sarcolaenaceae, (3) Thymelaeaceae (including Tepuianthaceae), (4) Muntingiaceae, and (5) Neuradaceae. The largest group within the order is Malvaceae s. l. (including Tiliaceae, Sterculiaceae, and Bombacaceae—Alverson et al., 1998 , 1999 ; Baum et al., 1998