Phylogeny of Agavaceae Based on ndhF , rbcL , and its Sequences

Great advances have been made in our understanding of the phylogeny and classification of Agavaceae in the last 20 years. In older systems Agavaceae were paraphyletic due to overemphasis of ovary position or habit. Discovery of a unique bimodal karyotype in Agave and Yucca eventually led to a reexamination of concepts and relationships in all the lilioid monocots, which continues to the present day. Developments in cytogenetics, microscopy, phylogenetic systematics, and most recently DNA technology have led to remarkable new insights. Large-scale rbcL sequence studies placed Agavaceae with the core Asparagales and identified closely related taxa. Analysis of cpDNA restriction sites, rbcL, and ITS nrDNA sequences all supported removal of Dracaenaceae, Nolinaceae, and clarified relationships. Agavaceae s.s. presently consists of Agave, Beschorneria, Furcraea, Hesperaloe, Hesperoyucca, Manfreda, Polianthes, Prochnyanthes, and Yucca. In this paper we analyze recently obtained ndhF sequence data from Agavaceae and Asparagales and discuss the implications for classification. Parsimony analysis of ndhF data alone resolves most genera of Agavaceae and supports the inclusion of Camassia, Chlorogalum, Hesperocallis, and Hosta within Agavaceae s.l. Analysis of combined ndhF and rbcL data sets of selected Asparagales results in better resolution and stronger bootstrap support for many relationships. Combination of all available ndhF, rbcL, and ITS data in a single analysis results in the best resolution currently available for Agavaceae s.l. Implications for classification schemes past and present are discussed.


INTRODUCTION
Agavaceae are a family of rosette-forming, often spiny plants centered in warm and dry areas of Mexico, the southwestern USA, and Caribbean basin.The family includes plants of great natural beauty, ornamental value, cultural significance, and economic importance.Although the family is easily recognized there have been questions about the circumscription of Agavaceae, how the genera are related to each other, and how Agavaceae are related to other families.In recent years, the family has been treated broadly with 18 genera (Cronquist 1981) and narrowly with nine core genera (Dahlgren eta!. 1985;Verhoek 1998).The undisputed genera of Agavaceae s.s. are Agave L., Beschorneria Kunth, Furcraea Vent., Hesperaloe Engelm., Hesperoyucca (Engelm.) Baker, Manfreda Salisb., Polianthes L., Prochnyanthes S. Watson, and Yucca L. Associated genera such as Camassia Lind!., Chlorogalum (Lind!.) Kunth, Hesperocallis A. Gray, and Hosta Tratt.have certain similarities to Agavaceae and have been considered closely related by some, but have never been fully integrated into the family.
on relatively few morphological characters to classify Agavaceae and its allies (Bentham and Hooker 1883;Engler 1888).In the 20'h century, comparisons of karyotypes (McKelvey and Sax 1933;Sat6 1935), embryology (Wunderlich 1950), anatomy and microcharacters (Huber 1969;Dahlgren and Clifford 1982), pollen (Alvarez and Kohler 1987), pollination and reproductive biology (Eguiarte 2000;Slauson 2001), and cladistic analysis (Dahlgren and Rasmussen 1983;Hernandez 1995a) have all contributed to our understanding of Agavaceae and their relationships.The greatest advances in the last 20 years have been made using molecular markers, especially DNA sequence data (Eguiarte et a!. 1994(Eguiarte et a!. , 2000;;Bogler andSimpson 1995, 1996;Chase eta!. 1995).Advances in DNA sequencing technology have made it relatively easy for botanists to acquire molecular sequence data.Increased computer power has enabled researchers to combine sequence data from different gene regions and analyze very large data sets indeed (Fay et a!. 2000;Soltis et a!. 2000).
In this paper we review some of the major ideas in Agavaceae classification and report the results of phylogenetic studies utilizing recently acquired chloroplast ndhF sequences, alone and in combination with previously published rbcL and nuclear ITS rDNA data sets.The results are integrated with past and present ideas concerning classification of Agavaceae.
Yucca was grouped with Dasylirion and Dracaena, whereas Agave, Beschorneria, Furcraea, and Polianthes went into Amaryllidaceae, subfamily Agavoideae.Lotsy (1911) was perhaps the first to refer to this group as Agavaceae, although still without Yucca.
Perhaps the greatest advance in the formal classification of Agavaceae came with Hutchinson's (1934) Families of Flowering Plants, in which Yucca was added to the family.In this system the xerophytic vegetative habit was considered as important as floral characters such as ovary position.Hutchinson's Agavaceae included Agave, Beschorneria, Furcraea, Hesperaloe, Manfreda, and Yucca, recognized as Agavaceae s.s. at the present time, and also Cordyline Comm.ex.R. Br., Dasylirion, Doryanthes Correa, Dracaena, Nolina Michx., Phormium J. R. Forst.& G. Forst., and Sansevieria Thunb.At about this time, the unusual bimodal N = 30 karyotypes of Agave and Yucca, with 5 large and 25 small chromosomes, were reported by McKelvey and Sax (1933), Whitaker (1934), andSato (1935).The strong resemblance of the Agavaceae karyotype to bimodal karyotypes of Camassia, Hesperocallis, and Hosta was also noted and the possibility that these genera might be related to Agavaceae was suggested by Granick ( 1944 ).
In recent times, Cronquist ( 1981) adapted the broad treatment of Hutchinson (1934) with few changes, even though he realized some of the genera were out of place.Cronquist was concerned that if he couldn't use habit to recognize families, then some other families would fall.Besides, large families such as Liliaceae are simple to teach and remember.In contrast, Takhtajan (1980) recognized many small families, and a narrowly circumscribed Agavaceae, but he included Hosta in the family.Influenced by advances in anatomy, phylogenetic systematics, and cladistic theory, Dahlgren et al. (1985) also recognized many small families, but made more of an effort to recognize only families that were demonstrably monophyletic.Putting the monophyletic pieces back together into larger groups has proven to be quite challenging.Using primarily microcharacters of the seed coat, cuticle, endosperm, and embryo, Huber (1969) aggregated many of these small families into larger groups, which he informally identified as asparagoid, dioscoreoid, and so forth.Agavaceae were placed in the asparagoid group, which was treated as order Asparagales by Dahlgren et al. ( 1985).Many Asparagales are characterized by berry fruits or capsules, and seeds encrusted by a black pigment known as phytomelan.
The earliest molecular studies involving Agavaceae were the immunological studies of Chupov and Kutiavina ( 1981 ).Using serology and immunoelectrophoresis techniques on a wide array of lilioid monocots, they demonstrated that both Camassia and Hosta had a very strong serological affinity to Agave and Yucca.Their results went largely unnoticed in the West, but supported the ordinal rank of Asparagales (Dahlgren 1983).The chloroplast gene rbcL has been very useful in studies of monocot phylogeny (Chase et al. 1995).These rbcL data strongly supported Asparagales as a distinct clade separate from Liliales.Two constant groups were recognized within Asparagales, a lower paraphyletic grade of families characterized by simultaneous microsporogenesis, and a higher monophyletic clade characterized by successive microsporogenesis.The rbcL data provided some support for potential family groups within Asparagales, but the bootstrap values were low and many branches collapsed in the strict consensus trees.The rbcL data showed that the family Agavaceae as circumscribed by Hutchinson (1934) and Cronquist ( 1981) was not monophyletic and that Agavaceae s.s. were on a separate clade from a Nolinaceae-Dracaenaceae clade and the other taxa (Eguiarte et a!. 1994 ).
Chloroplast DNA (cpDNA) restriction site studies (Bogler and Simpson 1995) strongly supported and expanded the hypothesis that Agavaceae s.s. were on a separate clade from the Dracaenaceae-Nolinaceae-Convallariaceae (sensu Dahlgren eta!. 1985) genera included in the family.Beaucarnea Lem., Calibanus Rose, Dasylirion, Dracaena, Nolina, and Sansevieria were all on a clade with berry-fruited, softleaved genera such as Aspidistra Ker Gawl., Maianthemum F. H. Wigg., and Polygonatum Mill., included in Liliaceae by Cronquist (1981 ).The cpDNA also provided resolution for the genera of Agavaceae.Hosta was shown to share molecular synapomorphies with Agavaceae.Hesperaloe and Yucca whipplei Torr.(= Hesperoyucca) were demonstrated to be closely related.Beschorneria and Furcraea were also closely related, and together formed a sister group to Agave s.l.Subsequent studies using ITS rDNA sequence (Bogler and Simpson 1996) confirmed most of these relationships, and provided additional resolution.The ITS data strongly indicated a close relationship of Camassia and Hosta to Agavaceae.Molecular variation within Agave was very low in these studies.
One of the more innovative approaches to higher level classification of angiosperms that takes into account results of molecular phylogenies is the Angiosperm Phylogeny Group (APG 1998;APG II 2003).In this system all taxa are monophyletic, paraphyletic taxa are not accepted.There are no subclasses, only informal higher groups (e.g., asterids, eudicots, magnoliids, monocots, rosids, etc.), orders, and families.The system is based almost entirely on molecular phylogenies, with contributions from many people.It is mostly concerned with establishing monophyletic orders that refer directly to large clades, with a secondary focus on families.The first edition (APG 1998) recognized 40 orders and 462 families.Asparagales are recognized as an order with 29 families, Agavaceae among them.The APG (1998) family nomenclature is illustrated in Fig. 1 and 2. A potential problem with using the criterion of monophyly is that it tends to result in rampant splitting and a proliferation of families, making it difficult to remember or teach families, and creates practical problems for floristic botanists.Nevertheless, APG offers the hope of a system that will stabilize as more phylogenetic data are incorporated.The APG II (2003) update recognizes 25 families in Asparagales and suggests ways of reducing the number of families if the user so chooses.The core Asparagales contain I 1 families, but can optionally be reduced to two families, Alliaceae and Asparagaceae, with Agavaceae going into the Asparagaceae (see Fig. 3).
In this paper we analyze the position of Agavaceae within Asparagales with new combinations of data, present evidence for including additional genera in Agavaceae, and draw together all the available sequence data into a single combined analysis of relationships of the genera of Agavaceae.

MATERIALS AND METHODS
The selection of taxa for this study was guided by our focus on family and generic relationships of Agavaceae and by the availability of sequence data.The sequences used in these analyses were obtained from a variety of sources (listed in Table 1).The rbcL sequences were all downloaded from GenBank.Many of the ndhF sequences were generated by J. Chris Pires in collaboration with Mark Chase, Sean Graham, and others using the protocol outlined in Pires and Sytsma (2000).Agavaceae ndhF sequences were generated by David Bogler at Florida International University and Fairchild Tropical Botanic Garden in the lab of Javier Francisco-Ortega.The ndhF sequences were amplified and sequenced in three overlapping pieces using various combinations of primers (listed in Table 2), derived from the paper by Terry et a!. ( 1997) on ndhF sequence variation in Bromeliaceae.
Standard cycle sequencing techniques were used with an ABI 377 (Applied Biosystems, Foster City, California, USA) automated sequencer at FlU, using standard reagents and the manufacturer's protocols.The ndhF sequences were combined and aligned using CLUSTALX (Thompson et al. 1997).The data sets were analyzed separately and in various combinations with PAUP* vers.4.0 (Swofford 1998).The rbcL sequences were all downloaded from GenBank and aligned using CLUSTAL_X.ITS 1 and ITS 2 sequences are the same ones used in a previous study of Agavaceae (Bogler and Simpson 1996).The ITS primers, sequencing technique, alignment procedures, GenBank accession numbers, and voucher specimens are provided in that paper.
Sequence data sets were assembled and combined using MacClade vers.3.05 (Maddison and Maddison 1992) and PAUP*.Sequences in the data sets were trimmed at each end so that they began and ended at the same character.Parsimony searches were made using heuristic settings for all analyses, with all characters unordered and equally weighted, and tree-bisection-reconnection (TBR) branch swapping in effect.Strict consensus trees were generated from the pool of most parsimonious trees found.Bootstrapping was carried out in PAUP* using 100 replicates with the number of trees saved per replicate limited to 100.
The first analysis examined sequence variation within the ndhF data set alone and its relative utility in resolving relationships of Agavaceae and selected Asparagales.Complete ndhF sequences were unambiguously aligned and analyzed with PAUP*.The ndhF data matrix included 38 taxa and a total of 286 parsimony informative characters.
For the second analysis, we focused on the relationships of Agavaceae within Asparagales using combined ndhF and rbcL data sets with a larger number of taxa representing major clades of Asparagales.In the majority of cases, the species were the same in both rbcL and ndhF data sets.In a few cases, the species in the rbcL data set differed from those in the ndhF data set, but were combined to represent the genus in the analysis.For a few taxa only partial ndhF sequences were available.The combined rbcL-ndhF data matrix included 60 taxa, with a total of 925 parsimony-informative characters.
In  three markers, ndhF, rbcL, and ITS.For some additional members of Agavaceae only partial ndhF sequences from the 3' region were available.ITS sequences were available only for Agavaceae.Inclusion of partial data is justified because both rbcL and the 5' region of ndhF are relatively conservative and exhibit little or no variation within Agavaceae s.s.(Chase et al. 1995;Eguiarte et al. 1994).ITS sequences are for the most part too variable to make reliable alignments outside Agavaceae s.s.(Bogler and Simpson 1996)    trnL-F (Fay et a!. 2000) in a large data set like this might actually decrease resolution and bootstrap support by introducing homoplasy due to misalignment or base saturation.The data matrix included 43 taxa and 578 parsimony informative characters.

Evaluation of ndhF Sequence Data
The relative utility of ndhF sequence data was assessed by parsimony analysis of complete ndhF sequences for 38 taxa of Agavaceae s.s. and selected core Asparagales.This data set included almost all genera of Agavaceae s.s. and basal Agavaceae s.!.-Camassia, Chlorogalum, He.lperocallis, and Hosta.This analysis resulted in 36 equally parsimonious trees, each with a consistency index of 0.655 and a retention index of 0.720.The strict consensus of these 36 trees is shown in Fig. 1.The tree is rooted using Agapanthus L'Her., based on the results of previous studies.In general, resolution and support is very good in the more terminal nodes, and rather weak in some of the lower nodes.The extent to which Agavaceae are resolved by ndhF variation is about the same, or perhaps a little less, than that achieved with combined ITS 1 and ITS 2 data (Bogler and Simpson 1996), which is about one-fourth the length of ndhF Aphyllanthes L.-Themidaceae are weakly supported as sister to Convallariaceae s.l.(following APG 1998, including Dracaenaceae, Nolinaceae, and Ruscaceae).Hyacinthaceae are on a strongly supported clade, and sister to Anemarrhena Bunge, Herreria Ruiz & Pav.-Behnia Didrichsen-Anthericaceae, and Agavaceae.The ndhF data strongly support Hesperocallis, Camassia-Chlorogalum, and Hosta as basal members of the Agavaceae clade (bootstrap 94% ), although resolution of these taxa relative to each other is weakly supported.Agavaceae s.s. are more or less resolved by the ndhF data, and comparable to the resolution by cpDNA restriction sites and ITS data (Bogler andSimpson 1995, 1996).Hesperaloe and Hesperoyucca are resolved as sister taxa, although these are not on the same branch with Yucca as in the ITS analysis (Bogler and Simpson 1996).Furcraea is strongly supported (95%) as sister to Agave s.!. (including Manfreda, Polianthes, and Prochnyanthes).Resolution within Agave is very low.Agave dasylirioides and A. striata differed from other species of Agave, but were not strongly united as they were with ITS data.This is surprising because they share at least two unique ndhF synapomorphies.The ndhF sequence of A. striata contained a unique 142-base deletion in the 5' region, not shared by A. dasylirioides or any other species in this small sample, and perhaps this deletion affects the resolution.Polianthes and Prochnyanthes are strongly united by several synapomorphies.

Combined ndhF and rbcL Data Sets
Analysis of combined ndhF and rbcL data sets for 60 selected taxa of Asparagales and Liliales resulted in 32 equally parsimonious trees, each with a length of 4431 steps, a consistency index of 0.453 and a retention index of 0.593.The strict consensus of these 32 trees is shown in Fig. 2. Overall, the topology is congruent with studies of rbcL data alone (Chase et a!. 1995), but combining ndhF with rbcL data resulted in greater resolution in the strict consensus tree and higher bootstrap support values.The topology is very similar to that obtained by Fay et a!. (2000) in an analysis of combined rbcL, atpB, trnL intron, and trnL-F sequences.The taxa representing Liliales form a strongly supported clade that was designated as the outgroup for presentation purposes, although broader studies have shown that the ordinal relationships are not yet completely resolved (Chase et a!. 2000).Asparagales are well supported by our data, as they are in studies of rbcL alone (Chase et a!. 1995) and in combination with other markers (Fay et a!. 2000;Soltis et a!. 2000).At the base of Asparagales is a paraphyletic grade of clades often referred to as lower Asparagales.At the very base are the orchids, an extremely large group here represented by Cypripedium L., and a clade including Blandfordia Sm., Curculigo Gaertn.and Hypoxis L., and Astelia Banks & Sol.ex R. Br. and Milligania Hook.f., in Asteliaceae.The positions of Doryanthes and Iris L. are congruent with previous studies, as is the clade containing Aloe and its relatives, and Xanthorrhoea Sm.Of interest here is the observation that some of these lower Asparagales with woody trunks and fibrous leaves were once placed in Agavaceae or considered closely related (Cronquist 1981 ).
The so-called higher or core Asparagales clade receives bootstrap support of 71%.At its base is a clade containing the Agapanthaceae-Amaryllidaceae-Alliaceae alliance, supported by bootstrap values of 93%, and congruent with previous studies of rbcL alone and in combination.This is followed by a large clade containing Asparagaceae, Convallariaceae, and Laxmanniaceae, using the APG (1998) names for families.Support for this clade is rather poor (bootstrap value less than 50%), indicating a need for additional attention.The branch uniting woody taxa such as Calibanus and Dracaena with the smaller, herbaceous taxa such as Maianthemum, Polygonatum, and Peliosanthes, is very strong, with 100% bootstrap support, but variation and resolution within this clade are low.The branch supporting the odd sub-Saharan genus Eriospermum Jacq.ex Willd.as sister to Convallariaceae s.l. has relatively weak (56%) bootstrap support in our analysis, in contrast to the core support reported by Fay et a~.(2000) and McPherson et a!.(submitted).
Agavaceae belongs to a complex of clades within the core Asparagales.The topology of this complex based on combined ndhF and rbcL data is congruent with previous analyses of rbcL data alone and in combination (Fay et a!. 2000).The data strongly supports the close relationship of Camassia, Chlorogalum, and Hosta to Agavaceae s.s., here represented by Polianthes.The sister group of Agavaceae is a clade containing Behnia, Herreria, and representative Anthericaceae, with 100% bootstrap support.Behnia and Herreria are basal to Anthericaceae, with 90% bootstrap support, but the relationship between these taxa is unresolved in this tree.Anthericaceae are supported by 100% bootstrap values.Anemarrhena is supported as sister to this whole Agavaceae-Anthericaceae complex.Hyacinthaceae are well supported with 100% bootstrap support, as is Themidaceae.In this study, Aphyllanthes is weakly supported at the base of Themidaceae, although in other studies it is weakly supported at the base of Hyacinthaceae (Fay et a!. 2000), sister to Laxmanniaceae (Pires et a!. 2006)

Combined ndhF, rbcL, and ITS Data Sets
For the third analysis, we combined all the available rbcL, ndhF, and ITS sequences of Agavaceae and associated core Asparagales into a single data set.This data included complete and partial sequences, and some taxa for which one or another sequence was entirely lacking, as indicated in Table 1.Parsimony analysis resulted in well over 1000 equally parsimonious trees, each with a length of 1985 steps, a consistency index of 0.691, and a retention index of 0.698.The strict consensus of 1000 equally parsimonious trees is presented in Fig. 3, rooted with Agapanthus.The large number of trees found in the heuristic search is probably related to the inclusion of species of Agave with nearly identical sequences, and perhaps to the incomplete nature of the data sets.The overall topology is very similar to the other two analyses presented in this paper, which is not surprising since they are based on many of the same sequences, but the representation of Agavaceae taxa is more complete and the resolution better than in previous studies.Some of the differences in resolution and bootstrap support are probably influenced by the stronger resolution of Agavaceae s.s.afforded by addition of the ITS data set.The position of Aphyllanthes is somewhat unstable, as seen elsewhere (Pires et al. 2006;McPherson et al. submitted).Here it is sister to the rest of the taxa included in this study.In the previous two analyses it was sister to Themidaceae, and sister to Hyacinthaceae in   Fay et al. (2000).Asparagaceae-Convallariaceae are strongly supported and sister to the remaining taxa.Themidaceae and Hyacinthaceae are sister taxa in this tree, though bootstrap support for this relationship is less than 50%.
In all three analyses presented here, Anemarrhena is sister to the group containing Agavaceae s.J. and Herreria-Behnia-Anthericaceae, with moderate bootstrap support of 75%.For this reason, Anemarrhena and taxa above this node might be considered as a well-supported, monophyletic, extended Agavaceae family as suggested by APG II (2003).This clade consists of two major groups with very strong bootstrap support.One group contains Behnia and Herreria, resolved in this analysis, and the strongly supported clade containing Anthericaceae.In all analyses this group is the sister to Agavaceae.The other strongly supported clade is Agavaceae s.J. with its associated basal genera.Hesperocallis, for which only ndhF data was available, is strongly supported as a basal member of Agavaceae, with 97% bootstrap support.As in all previous molecular studies, Camassia, Chlorogalum, and Hosta are placed somewhere near the base of Agavaceae.The relative position of Hosta and Chlorogalum-Camassia is variable in these analyses and bootstrap support values are low in this part of the tree, indicating the need for additional study.
Hesperaloe and Hesperoyucca are strongly supported as sister taxa, congruent with previous cpDNA and ITS sequence studies (Bogler andSimpson 1995, 1996;Bogler et al. 1995).This pair comes out as basal to the rest of Agavaceae s.s., but with low bootstrap support.This position is similar to the tree derived from cpDNA restriction site analysis (Bogler and Simpson 1995), but in the combined ITS data tree Hesperaloe and He.1peroyucca are on a separate clade that includes Yucca and Camassia.At this point, it is difficult to say whether this difference is influenced by the amount or quality of data, or perhaps an artifact of combining partial data.The three species of Yucca are strongly supported.Agavaceae taxa with an inferior ovary comprise a very strongly supported clade with a bootstrap value of 100%.Beschorneria and Furcraea are joined, however, only ITS data was available for Beschorneria.
Agave s.l.falls with in the Agavaceae clade, with moderate bootstrap support of 72%.Agave dasylirioides and A. striata appear on a separate branch with 96% bootstrap support.This support is higher than it was in the analysis of ITS sequences alone, probably reflecting the contribution of ndhF support.There is very little variation in ndhF or ITS sequences in the other species of Agave and no resolution.Several species had identical sequences, but were included anyway to emphasize the lack of variation.The only taxa that are consistently resolved are the two species of Polianthes and Prochnyanthes mexicana, which share a few unique ndhF and ITS characters.Manfreda virginica was not resolved from Agave in this analysis, though cpDNA linked M. virginica to A. lecheguilla (Bogler and Simpson 1995) and ITS data alone linked it to Polianthes-Prochnyanthes (Bogler and Simpson 1996).The lack of sequence variation in Agave is frustrating, and has resulted in one of us screening cpDNA spacer sequence regions for variation (atpB-rbcL, trnT-trnL, trnH-psbA, rpl20-rpsl2, psbB-psbF), but with limited results (D. Bogler unpubl.data).Current efforts are directed toward using DNA fingerprinting techniques to study the groups of Agave recognized by Gentry ( 1982).

Asparagales
This study has provided insights into the phylogeny of Agavaceae at several levels.The overall topology of the Asparagales tree derived from combined ndhF and rbcL data sets (Fig. 2) is similar to those of other investigators (Chase et al. 1995;Fay et al. 2000).Lilioid monocots can be divided into several large groups that have been formerly named as orders, and Agavaceae clearly belong in the group known as Asparagales.All Agavaceae have the black phytomelanencrusted seeds characteristic of Asparagales.Most Asparagales are hypogynous, but epigyny has evolved independently several times, as in the case of Agave.Our study supports the division of Asparagales into two parts, a paraphyletic grade of clades, and a large, monophyletic, clade.Although these have been referred to as lower Asparagales and higher Asparagales, we reject any notion of progress these terms may imply, and prefer to use the term core Asparagales for the monophyletic clade.A detailed discussion of the lower Asparagales grade is beyond the scope of this paper and can be found elsewhere (Fay et al. 2000;Reveal and Pires 2002;Graham et al. 2006;Pires et al. 2006).Here we only note that several of the woody taxa with fibrous leaves were once placed in or near Agavaceae in older systems of classification.Hutchinson (1934) included Cordyline, Doryanthes, and Phormium in Agavaceae.Cronquist (1981) placed Aloe and Xanthorrhoea in nearby families he considered "parallel derivatives from the Liliaceae," which in a sense they really are.But the molecular data clearly show that Aloe and Xanthorrhoea are rather distantly related to Agavaceae s.s.(Bogler and Simpson 1996).An interesting corollary to this finding is that woody habit and fibrous leaves have arisen independently in a number of Asparagales in different parts of the world, possibly in response to herbivory or fire regime.

Core Asparagales
It is worthwhile to consider the core Asparagales in more detail.Sister group to the core Asparagales is a strongly supported clade that contains Agapanthaceae, Alliaceae, and Amaryllidaceae (following APG 1998) (Fig. 2).These families are generally characterized by having bulbs, scapes, and umbels subtended by bracts.Agapanthus and Allium L. are hypogynous and taxa of Amaryllidaceae are epigynous, a situation that parallels that of Agave and Yucca and has caused similar problems in classification.These taxa have been classified as tribes of Liliaceae (Krause 1930) or, after certain disparate members have been removed, as separate families (Dahlgren et al. 1985;Kubitzki 1998a;Meerow and Snijman 1998;Rahn 1998).The phylogeny of these families has been extensively studied by Meerow et al. (2000).The first APG ( 1998) treatment listed these as separate families, but the second APG II ( 2003) update suggests it might be easier to optionally treat all three as the single family Alliaceae, recognized by the umbellate inflorescence subtended by two external bracts.In a similar fashion, the remaining families of core Asparagales might be reduced to the single family Asparagaceae, characterized by racemose inflorescences (see Fig. 3).
The other large group of core Asparagales contains the Agavaceae group, Hyacinthaceae, and Themidaceae.Somewhat isolated here is the monotypic genus Aphyllanthes, whose sole species occurs in the western Mediterranean region (Dahlgren et al. 1985;Conran 1998b).The leaves of this odd plant are reduced to scales and the photosynthetic organs are the scapes.The flowers are borne at the ends of the scapes in small, condensed spikes surrounded by scarious bracts.The fruits are capsules with a few black seeds.Based on morphology, it has been compared to Eriocaulaceae or Xanthorrhoeaceae, but all molecular analyses place Aphyllanthes with the core Asparagales.The ndhF data place Aphyllanthes near Themidaceae, but with low support.Given these anomalies, Aphyllanthes would be a good candidate for additional sampling and sequence verification.Themidaceae were once included in Alliaceae, but were "resurrected" as a distinct family on the basis of rbcL sequence studies (Fay and Chase 1996).Our data supports the separation of Themidaceae from Alliaceae, and the combined data set analysis indicates a sister relationship to Hyacinthaceae, but with bootstrap support of less than 50%.Hyacinthaceae are characterized by having bulbs, leafless scapes, and racemes or spikes, but not umbels as in Alliaceae.Camassia, Chlorogalum, and several other genera formerly placed in Hyacinthaceae (Speta 1998) are here shown to be basal members of Agavaceae.

Agavaceae Sensu Latissimo
The final clade to consider is the one containing all families and genera closely associated with Agavaceae in the broadest sense.It is not yet possible to identify unique morphological synapomorphies for this group.A lot of attention has been given to the karyotypes, since these have proven to be useful indicators of relationship to Agavaceae.The molecular data place Anemarrhena at the base of this clade, with moderately strong bootstrap support of 75% in the combined data trees.Anemarrhena is a monotypic genus from northern China and Korea.It is an understory herb, with a rhizome and short stem, linear, sheathing leaves, a pedunculate, compact panicle, mostly free tepals, three stamens attached to the tepals, superior ovary, and an ovoid capsule with one or two black seeds (Conran and Rudall 1998).The chromosome number is n = 11.The unusual reduction in stamen number has been used to associate Anemarrhena with Hemiphylacus and Johnsonia R. Br., but the molecular data do not support those relationships.
There is very strong bootstrap support (Fig. I) for the sister group status of Herreriaceae-Behniaceae-Anthericaceae and Agavaceae s.l., with all the basal genera included.Both ndhF and rbcL strongly support this relationship.At the base of this clade are Behnia and Herreria, weakly resolved in our combined analysis.Behnia is a genus with a single species, B. reticulata from moist, shady forests in southern Africa (Conran 1998c).They are dioecious, slender, shrubby plants from short rhizomes with secondary thickening, thin erect branching stems, and shortly petiolate ovate leaves.The flowers are borne in small axillary cymose clusters or singly, campanulate, with partially united tepals, and six stamens adnate to the tepals.The fruit is a pale yellow or white berry with few angular, seeds; dark, but not phytomelan encrusted.With its broad leaves, campanulate flowers, berries, and lack of phytomelan, Behnia has been related to a variety of families including Asparagaceae, Luzuriagaceae, Ruscaceae, and Smilacaceae (Dahlgren et al. 1985;Conran 1998c).Of all the taxa that appear to be related to Agavaceae, Behnia is the most difficult to rationalize in terms of morphology, and is also a good candidate for independent verification.
Herreria is a genus of eight species in subtropical and temperate South America (Dahlgren et al. 1985;Conran 1998d).They are understory shrubs, acaulescent or caulescent with branching prickly stems several meters tall and secondary thickening.Leaves are linear to lanceolate, often clustered on short lateral shoots.Flowers are borne in small, scaly panicles.Flowers are hypogynous and bisexual.Fruits are trilobed thin-walled capsules, with flattened winged black seeds.The chromosome number is n = 27, with one large chromosome and 26 short ones (Sat6 1942).This bimodal karyotype has been taken as a potential link to Agavaceae.There is another genus of Herreriaceae, Herreriopsis Perrier in Madagascar, which differs in having sac-like tepa! bases and numerous ovules per locule.Dahlgren et a!.(1985) thought that Herreria resembled Asparagaceae or Ruscaceae, but the molecular data indicates they are not closely related.
Anthericaceae are strongly supported as a clade.As circumscribed by Conran (1998a), Anthericaceae s.s. are a family of 9 genera, about 200 species, and worldwide in distribution.They are somewhat herbaceous, from short rhizomes, with fibrous, fleshy or tuberous roots.The leaves are narrow, sheathing, and spirally arranged.Flowers are borne in erect scapose racemes or cymes.The flowers are bisexual, hypogynous, with mostly free tepals.The fruit is a dry elongated capsule with black angular somewhat flattened seeds (Conran 1998a).Chromosome numbers in the family are quite variable, with polyploidy and aneuploidy found in some taxa (Conran 1998a).In older systems Anthericaceae was a much larger family and included members of what are now classified as Boryaceae, Johnsoniaceae, and Lomandraceae.Dahlgren et a!.(1985) tentatively positioned Anthericaceae near Asphodelaceae.Leucocrinum montanum, from western North America, has been associated with Hosta (Dahlgren eta!. 1985) and with Hemiphylacus (Hernandez 1995b).The flowers of Leucocrinum Nutt.ex A. Gray are tubular, with the ovary and capsule located below ground level.Chlorophytum comosum is a common houseplant called the "spider plant" because of the long, spreading inflorescence stalks.Echeandia occurs primarily in the New World and Anthericum L. in the Old World, but the generic limits of these two genera are uncertain.

Early Diverging Members of Agavaceae
The molecular studies described here provide strong support for the inclusion of Camassia, Chlorogalum, Hesperacallis, and Hosta in Agavaceae.Although these genera were known to have features in common with Agavaceae, they were usually placed in other families such as Hyacinthaceae because of their more generalized "liliaceous" appearance.These "basal Agavaceae" seem to have proliferated mostly in western North America, which is probably the area of origin of the entire clade.The ndhF data supports the position of Hesperocallis at the base of Agavaceae s.l.The single species, H. undulata, occurs in the Sonoran Desert.The leaves are narrow, linear, folded, with undulate margins, and are borne in a rosette from a small corm-like rhizome.The inflorescence is a bracteate raceme, with the flowers tending to be on one side.The tepals are partially united, forming a tube that withers and persists after flowering.The fruit is a three-lobed subglobose capsule, with fiat black horizontally arranged seeds.The chromosome number is n = 24, with 5-6 very large and 19-18 small chromosomes (Kubitzki 1998b ).Cave ( 1948Cave ( , 1970) ) discussed similarities between embryology and karyotypes of Hesperocallis and Agavaceae.In the past, the classification of Hesperocallis was generally uncertain.Krause (1930) placed Hesperocallis in tribe Hemerocallideae, with Blandfordia, Hemerocallis L., Hosta, Leucocrinum, and Phormium.Hutchinson (1934) included only Hemerocallis, Hesperocallis, Hosta, and Leucocrinum.The ndhF support for inclusion of Hesperocallis in Agavaceae is very strong, but additional sequence markers are needed to confirm the relationships among the early diverging genera.
Hosta is a genus of 23-50 species native to shady, temperate forests in China, Japan, and Korea.It is a popular ornamental plant and there are many cultivars.Hosta has clumpy rhizomes with fleshy roots.The leaves vary in shape from linear or oblong to lanceolate or broadly ovate, and sometimes are narrowed at the base forming a pseudopetiole.The inflorescence is an elongate bracteate raceme, with the flowers often on one side as in Hesperocallis.The flowers are hypogynous with a tubular perianth and six stamens.The fruit is an oblong to subglobose capsule with black flattened winged seeds, similar to those of Agave.Steroidal saponins are present, as in Agavaceae.The embryology and pollen resemble Hesperocallis (Kubitzki l998b).The chromosome number is n = 30 for most species, with 4 large, 2 medium, and 24 small chromosomes.Triploids and aneuploids also are known.McKelvey and Sax (1933), Whitaker (1934), andSat6 (1935) noted the similarity between the karyotypes of Hosta and Agavaceae many years ago.Granick (1944) suggested that Hosta was an early offshoot of the lineage leading to Agavaceae s.s., which appears to be correct, and that it must have been more widespread in the past, which is uncertain.It is interesting to speculate that the ancestor of Hosta might have originated in the New World and extended to China and Korea when there was a direct land connection.
The molecular data also support inclusion of Camassia and Chlorogalum as members of Agavaceae.These genera were formerly associated with subfamily Scilloideae of Liliaceae (Engler 1888;Krause 1930), and more recently included in Hyacinthaceae (Dahlgren eta!. 1985;Speta 1998).Camassia has six species, mostly in western North America, with C. scilloides widespread in the central USA (Gould 1942).Chlorogalum has five species in western North America, mostly in California (Hoover 1940).Camassia and Chlorogalum both have bulbs, characteristic of Hyacinthaceae, but some also have a short rhizome.A densely fibrous tunic, similar to Polianthes and Prochnyanthes, surrounds the bulbs.The leaves are narrow, and sometimes keeled.The inflorescence is a slender raceme in Camassia or a branched raceme or panicle in Chlorogalum.The tepals are separate and persist in fruit.The stamens are inserted at the base of the perianth.The fruit is a subglobose capsule with 2-12 pyriform rugose black seeds.Chlorogalum has 2 seeds per locule, but Camassia has 10-12 seeds per locule.Steroidal saponins are especially abundant in Chlorogalum, with some species known as "soap plant."The chromosome number reported for Camassia is n = 15, bimodal with 3 large chromosomes and 12 small ones (Gould 1942;Fernandez and Davina 1990).The chromosome numbers reported for Chlorogalum are n = 15, 17, 18, 30 (Cave 1970).The karyotypes of Chlorogalum are also bimodal (Cave 1970).In the n = l 5, 17, and 18 populations there are 3 chromosomes larger than the others, and in the n = 30 species there are 5 chromosome larger than the others.Cave (1970) thought the karyotype data was good evidence for keeping Camassia and Chlorogalum together in Tribe Scilleae in Liliaceae.We think it is good evidence for a very close relationship to Agavaceae.The bimodal n = 30 karyotype of Agavaceae might have originated by a doubling of a bimodal n = 15 genome in the ancestral species, which may have looked something like Camassia, Chlorogalum, or Hosta.
Any discussion of Camassia, Chlorogalum and other basal Agavaceae must also consider several similar genera for which molecular data are not yet available.Schoenolirion Torr. is a genus of three species in the southeastern USA.It is similar to Camassia, but has a short vertical rhizome below the bulb, 1-2 seeds per locule, a more congested inflorescence, and persistent perianth that becomes twisted in fruit.Hastingsia S. Watson has four species in California and southern Oregon.It was formerly included in Schoenolirion, but has a shorter pedicel and dimorphic stamens.The chromosome number reported for Hastingsia is n = 26 (Cave 1970), bimodal with 5-6 large chromosomes and 21-20 smaller ones.Fortunatia 1. F. Macbr. is a genus with four species in western South America that is also similar to Camassia.Cocucci ( 1969) thought it should be included in Camassia, however, Fernandez and Davina (1990) thought Fortunatia should be retained as separate, but Speta (1998) refers to it as the genus Oziroe Raf. and places it in a different subfamily than Camassia.The karyotype of Fortunatia is reported as n = 15 for three species by Fernandez and Davina (1990); bimodal with 3 large chromosomes and 12 small ones, and n = 17 for one species.Hopefully, all of these basal genera will some day be integrated into the phylogenetic analyses.

Agavaceae Sensu Stricto
The final group to discuss is Agavaceae s.s.Much of the background material on the phylogeny and morphology has been discussed and summarized elsewhere (Bogler andSimpson 1995, 1996;Verhoek 1998;Eguiarte et al. 2000).The basic outline of the tree derived from ndhF is similar to those derived from cpDNA and ITS data sets.The genera with a superior ovary, Hesperoyucca, Hesperaloe, and Yucca, are near the base of the clade, and the genera with an inferior ovary, Beschorneria, Furcraea, and Agave s.l., are united at the tip.In all molecular analyses Hesperaloe and Hesperoyucca are united as sister taxa with very strong support (Bogler et al. 1995;Bogler andSimpson 1995, 1996).This clade is extraordinarily diverse in pollination syndromes.Hesperaloe is a genus of 5-7 species in northern Mexico and south Texas with acaulescent rosettes of stiff, hard-fibrous leaves.The inflorescence is a terminal raceme or panicle from the rosette with flowers borne on small lateral spurs.Flower shape varies from tubular to campanulate, and are green, white, red, purple, pink, or yellow.Stamens are included or shortly exserted.The filaments are not swollen, and the style is filiform with a capitate, fringed stigma.The fruit is a short woody capsule with thin black seeds (Starr 1997).Hesperaloe species are pollinated by hummingbirds, hawkmoths, bees, and bats, and exhibit a wide range of associated floral forms, but have very low variation in molecular analyses.This would seem to indicate a very recent radiation of forms.
Hesperoyucca has one or two species in California, Arizona, and northwestern Mexico (Clary 2002).It has dense rosettes of hard-fibrous, spiny-tipped leaves.The inflores-cence is a branched panicle borne on a massive, bracteate stalk.Hesperoyucca is monocarpic, flowering once and then dying.The flowers are campanulate with separate, thick, creamy-white tepals.Superficially, the flowers of Hesperoyucca resemble those of Yucca and are pollinated by a species of Tegeticula, the genus of yucca moths that pollinate all species of Yucca.The filaments are swollen as in Yucca flowers, which act as a support for the moths as they crawl around.The style is short with a capitate papillose stigma like Hesperaloe, not like the sunken cavity as in Yucca flowers.The fruits are short capsules with flat black seeds.The strong molecular evidence linking Hesperaloe and Hesperoyucca implies that Hesperoyucca evolved yucca-like floral traits involved in pollination following colonization by the yucca moth (Bogler et al. 1995).
Yucca is a large genus of about 40 species primarily distributed in arid regions of Mexico and the southwestern USA.They vary from acaulescent grass-like shrubs to massive, strangely branched tree-like forms.The leaves are hard fibrous, sometimes spiny tipped, and borne in dense rosettes.The inflorescence is a bracteate raceme or panicle.The tepals are mostly separate, fleshy and white.The filaments are swollen.The ovary is superior, with three stigma lobes that form a sunken cavity.The fruits are dry or fleshy oblong capsules with flat black seeds.The pollination of Yucca involves a famous and fine-tuned mutualistic association with several species of Tegeticula, the yucca moth genus (Baker 1986).The fertilized Tegeticula female enters a Yucca flower and gathers a mass of pollen with specialized tentacles.She then flies to another Yucca flower, oviposits her eggs in the ovary, and places the ball of pollen in the sunken cavity formed by the stigma.The developing larvae feed on some of the seeds before crawling out of the fruit and dropping to the ground to pupate.Each section of Yucca is pollinated by a different species or species complex of Tegeticula, as is Hesperoyucca whipplei, which was classified with Yucca in the past.Yucca s.s.appears to be monophyletic.In the ITS analyses (Bogler and Simpson 1996), Yucca appeared on a clade with Camassia, Hesperaloe, and Hesperoyucca, but in the more conservative cpDNA analysis (Bogler and Simpson 1995) and in the ndhF and combined analyses presented here, Yucca is resolved as a separate clade.
Agavaceae genera with an inferior ovary form a strongly supported clade in all analyses.Beschorneria and Furcraea are united by the ITS and cpDNA data, but ndhF sequences were obtained only for Furcraea, and independent confirmation of this relationship is desirable.Fun•raea is a genus of about seven or eight species, widely distributed in the Antilles, Mexico, Central America, and in the Andes in South America (Garcia-Mendoza 2000).Furcraea varies in habit from acaulescent to forming massive trunks several meters tall.Leaves are linear to narrowly lanceolate, and sometimes with a spiny margin.Fun•raea is monocarpic, flowering once and then dying.The inflorescence is a massive, widely branched panicle.Flowers are pendulous, campanulate, with greenish or white tepals, an inferior ovary, and basally swollen filaments.Many species of Furcraea produce bulbils, and may depend on them for propagation.Beschorneria is a genus of seven species, all occurring in Mexico and Guatemala (Garcia-Mendoza 1987).Beschorneria is generally acaulescent or less commonly caulescent, with a rosette of rather succulent, linear-lanceolate leaves.The inflorescence is a somewhat lax few-flowered raceme with brightly colored bracts.The flowers are pendulous, red to greenish yellow, with separate tepals forming a tube, somewhat swollen filaments, an inferior ovary, and short capsule.Both Beschorneria and Furcraea produce pollen in tetrads.
The apex of the Agavaceae clade consists of Agave s.l., which contains Agave and subgroups that are sometimes recognized at the genus level such as Manfreda, Polianthes, and Prochnyanthes.Agave itself is a very large genus of about 200 species centered in Mexico, but widely distributed in the southwestern USA, Caribbean, Central America, and northern South America (Gentry 1982;Garcia-Mendoza 2002).Most species of Agave s.s. are acaulescent xerophytic rosette plants, with only a few developing weak trunks.The fibrous leaves vary greatly in size, succulence, and development of marginal spines.The inflorescence is a small to large spike-like raceme or spreading panicle with flowers in broad subumbellate clusters, which is the basis for division into the two subgen.Agave and Littaea.Most species are monocarpic, the rosette flowering once then dying, but a few are polycarpic, flowering year after year.The flowers of Agave s.s. are borne in pairs, with a yellow, tubular, or campanulate perianth formed by six equal or unequal tepals.The stamens are exserted with slender filaments, and the inferior ovary has a long, exserted style.Most species are protandrous and pollinated by a variety of bats, bees, and hummingbirds (Verhoek 1998;Eguiarte et a!. 2000;Slauson 2001).The fruit is a hard capsule with numerous fiat black seeds.Gentry (1982) recognized 20 groups of Agave in continental North America, but the relationships of these groups are obscure and in need of further study.The molecular data seem to indicate the subgenera may be paraphyletic.
Several genera are often recognized within Agave s.l.(Rose 1899(Rose , 1903(Rose , 1906)).Manfreda is a genus of about 26 species in the southeastern USA, Texas, Mexico, and Central America (Verhoek 1998), distinguished by thin or succulent, often mottled leaves and spike-like racemes with single flowers at the nodes.The tubular flowers are usually greenish yellow, white or pinkish, and have strongly exserted stamens and styles.Polianthes is a genus of about 14 species from Mexico (Verhoek 1998), distinguished by thin narrow leaves, flowers paired at the nodes, and stamens included in the tube.The flowers are white, pink, or red, sometimes strongly fragrant, and pollinated by moths and hummingbirds.Rose (1899) united Bravoa La Llave & Lex. with Polianthes.Prochnyanthes consists of a single somewhat variable species, P. mexicana, also from Mexico.It has thin, papery leaves, a lax raceme with paired, pink to greenish flowers that are often on a long pedicel, a greatly expanded floral tube, bent downward near the middle, included stamens, and is pollinated by sphinx moths and possibly bees.All of these genera have been included within Agave at various times in the past and sometimes in the present as well (Eggli 2001 ).
Molecular variation within Agave s.l. is rather low, but some groups were resolved in this small sampling.The ITS analysis (Bogler and Simpson 1996) provided some support for a basal clade containing A. dasylirioides and A. striata.Both species belong to Group Striatae as described by Gen-try (1982).This group was considered to have a number of unusual features that might be considered generalized or primitive within the genus (Gentry 1982).They are perennial and polycarpic, flowering repeatedly.The leaves are hard, and entire or finely serrate on the margin.The inflorescence is a simple spike, with paired flowers, nondimorphic tepals, an incompletely inferior ovary, and lack vegetative reproduction such as bulbils.Therefore, we think it is very significant to find the molecular data supporting the position of this clade at the base of the genus.The ndhF data for this relationship by itself were not quite as strong, but in the combined analysis the bootstrap support for this clade is strengthened.The other species of Agave s.s. were either not resolved by the ndhF data or weakly resolved into a single group by the ITS data.There is also both ndhF and ITS evidence for a close relationship between Polianthes and Prochnyanthes, which is not so surprising because of their many similarities.Manfreda is not resolved by the ndhF data, and there is only weak ITS support for a relationship to Polianthes-Prochnyanthes.There is simply not enough variation and too few taxa sampled to make any strong conclusions about relationships for most species.However, the finding of a clade supporting A. dasylirioides-A.striata raises some issues.If one chooses to recognize Manfreda, Polianthes, and Prochnyanthes as genera, then Agave becomes paraphyletic.The problem is solved by recognizing the entire clade as Agave, as was recently done by Eggli (2001), or by naming some additional genera.At present, it is probably premature to name new genera, but when more molecular data become available for the groups of Agave, this may be an attractive option.

CONCLUSIONS
These analyses have demonstrated the utility of ndhF sequence in resolving the relationships of Agavaceae.The advantage of using ndhF is that it has a relatively conservative 5' region and a more variable 3' region, which makes it well suited for large-scale studies and combination with other data sets.Analysis of ndhF alone provides support for inclusion of additional basal genera in Agavaceae and good overall resolution of the genera (Fig. 1).When ndhF is combined with rbcL data there is better resolution and stronger branch support for family level relationships within Asparagales (Fig. 2).When ndhF is combined with both rbcL and ITS data sets, the resulting tree is probably the best overall estimate of the phylogeny of Agavaceae currently available (Fig. 3).We think the addition of more taxa and sequence data will improve the resolution and stability of the relationships seen here.Adapting the nomenclature from previous systems of classification has not proven to be very easy, but with time this situation should also improve.
Fig. I.-Strict consensus of 36 equally parsimonious trees from analysis of the ndhF data set for Agavaceae and related genera.Family names follow APG (1998).Numbers at nodes are bootstrap support values.Branch lengths correspond to distance.

Fig
Fig. 2.-Strict consensus of 32 equally parsimonious trees found in analysis of combined ndhF and rbcL data sets from Asparagales and Liliales.Family names follow APG (1998).Numbers at nodes are bootstrap support values.Branch lengths correspond to distance.

100.
Fig. 3.-Strict consensus of equally most parsimonious trees found in analysis of combined ndhF, rbcL, and ITS data sets for Agavaceae.Family names follow APG II (2003), and illustrate the different options for recognizing family groups in Agavaceae and core Asparagales.Numbers at the base of the branches are bootstrap support values.Branch lengths correspond to distance.
for substantial bias in maximum parsimony-based inference of the phylogenetic position of Aphyllanthaceae (McPherson et al. submitted).
. Including ITS or other variable regions such as ). Numbers at nodes are bootstrap support values.Branch lengths correspond to distance.

Table I .
or sister to Agavaceae (McPherson et a!.submitted).Monte Carlo simulations have List of taxa included in sequence analyses and available GenBank accession numbers.