Molecular Studies of Subfamily Gilliesioideae ( Alliaceae )

We present an analysis of relationships in Gilliesioideae (Alliaceae) based on a combined matrix of plastid rbcL, the trnL intron, the trnL-F intergenic spacer, and the rpsl6 intron and nuclear ITS ribosomal DNA sequences. The results are generally congruent with previous analyses, indicating two well-supported groups: Ipheion plus allied genera ("lpheieae" ined.) and Gilliesieae. They also provide higher bootstrap support for many patterns of relationships. Polyphyly of lpheion and Nothoscordum is confirmed. Increased taxon sampling (particularly in Gilliesieae) and additional molecular data would be desirable to provide further resolution and to allow an appropriate taxonomic revision to be made.


INTRODUCTION
Alliaceae Batsch ex Borkh.are a family of Asparagales Bromhead sensu Angiosperm Phylogeny Group (APG;1998).They consist of ca. 14 genera and 600 species of bulbous or rhizomatous geophytes characterized by an umbel-like inflorescence subtended by two spathe-like bracts (sometimes fused), a superior ovary, a solid style, and the frequent occurrence of alliaceous chemistry.Alliaceae are most closely related to Amaryllidaceae J. St.-Hil.and the monogeneric Agapanthaceae F. Voigt.In APG II (2003) these three families were combined, with the three family names being "bracketed" under the oldest, Alliaceae.For bracketed names, APG II permits the recognition of the smaller units; here we use Alliaceae in the narrow sense (as in APG 1998) to allow more accurate descriptions of the clades.
Agapanthaceae have frequently been included in Alliaceae (e.g., Dahlgren et al. 1985), but phylogenetic analyses of molecular data from rbcL (Fay and Chase 1996), rbcL plus the trnL intron and the trnL-F intergenic spacer (together called trnL-F in this paper) (Meerow et al. 1999) and rbcL, trnL-F, and atpB (Fay et al. 2000) have demonstrated that Agapanthus L'Her.falls outside Alliaceae.Fay and Chase (1996) transferred Agapanthus to Amaryllidaceae, with which it was placed in analyses of rbcL data alone.However, in combined molecular analyses (Fay et al. 2000), although the clade consisting of Alliaceae, Amaryllidaceae, and Agapanthus was strongly supported, and each of these three groups was also strongly supported, the relationships between them were not resolved, with Agapanthus placed as sister to either Amaryllidaceae or Alliaceae plus Amaryllidaceae.For these reasons, and because Agapanthus Jacks some highly characteristic features of Alliaceae, such as the possession of a solid style (Rudall et al. 2002), we here exclude Agapanthus from Alliaceae.
The analyses of Fay and Chase (1996), Fay et al. (2000) and Pires et al. (200 1) also indicated that the cormous taxa previously often treated as tribe Brodiaeeae Traub of Alliaceae, e.g., by Dahlgren et al. (1985), were more closely related to other asparagalean families, including Hyacintha-ceae Batsch ex Borkh.On the basis of these results, Fay and Chase (1996) resurrected Themidaceae Salisb.for this group.A suite of morphological and anatomical characters, including corms vs. bulbs, absence of membraneous sheathing leaf bases, three or more spreading bracts vs. a spathaceous involucre of two (sometimes fused) bracts, and hollow style vs. solid style, add support to this separation of Themidaceae from Alliaceae (e.g., Hoover 1939).Themidaceae have been recognized in most subsequent treatments (Rahn 1998a, b;APG 1998;Thorne 2000;Chase et al. 2000).In APG II (2003), Themidaceae were bracketed under Asparagaceae Juss.s.l.together with six other families.
On account of its alliaceous odor the enigmatic genus Hesperocallis A. Gray has been included either in Alliaceae (e.g., Traub 1968) or in a monotypic family Hesperocallidaceae Traub allied to Alliaceae (Traub 1972b(Traub , 1982)).However, other authors (e.g., Cave 1948Cave , 1970;;Dahlgren et al. 1985) have suggested an affinity with Agavaceae Dumort.sensu APG (1998) or Hemerocallidaceae R. Br. (Hutchinson 1934(Hutchinson , 1959)).Molecular data have recently indicated a close relationship with Agave L. (Pires et al. 2004), and Hesperocallis is thus a member of Agavaceae or Asparagaceae sensu APG II (2003).Fay and Chase (1996) (including Nectaroscordum Lindl., Caloscordum Herb., and Milula Prain) and are sister to the other two subfamilies (Fay and Chase 1996;Fay et al. 1997;Meerow et al. 1999: Fay et al. 2000).Allium dregeanum Kunth from South Africa is the only member of the subfamily native to the southern hemisphere (e.g., De Sarker et al. 1997).Members of this subfamily differ from other Alliaceae in the possession of a gynobasic style (e.g., Di Fulvio 1973;Rudall et al. 2002).Nectaroscordum has been segregated from Allium by many authors on the basis of a semi-inferior ovary, but some (e.g., Traub 1972a) have included it in Allium.Molecular analyses of plastid DNA (Fay and Chase 1996;Friesen et a!. 2000) and the internal transcribed spacer (ITS) of nuclear ribosomal DNA (Dubouzet and Shinoda 1998) have shown Nectaroscordum to be embedded within Allium.Caloscordum has been treated as a distinct genus on the basis of the shape of its floral tube, the deeply inserted filaments without membranous appendages, a withering style, recurving lobes of the perianth, and a lack of alliaceous chemistry (Herbert 1844;Lindley 1847).Caloscordum has been included in Nothoscordum, but anatomy indicates a relationship with Allium rather than to Nothoscordum as the style is gynobasic (Lindley 1847;Hanelt eta!. 1991).Other authors have included Caloscordum in Allium (e.g., Li et a!. 1996), and molecular studies supported this relationship (Dubouzet and Shinoda 1998).Molecular studies have also shown Milula, which possesses a dense raceme instead of the characteristic umbel-like inflorescence, to be embedded in Allium (Friesen et a!. 2000).
Gilliesioideae, comprising all New World genera except Allium, are bulbous herbs.All genera are exclusively South American, with a concentration in southern Chile and Argentina, except Nothoscordum, which is widespread in South America but has some species that extend to North America.Most authors (e.g., Fay and Chase 1996;Rudall eta!. 2002) have recognized two groups within Gilliesioideae: (1) a group with actinomorphic flowers, sometimes referred to as tribe Ipheieae although this name does not appear to have been published, including Ipheion Raf., Nothoscordum, and Leucocoryne s.l.(sensu Rahn 1998a; including Pabellonia Quezada & Martie.and Stemmatium Phil.sampled here); and (2) tribe Gilliesieae Lindl.(Gilliesia Lindl.and allies) with zygomorphic flowers.Gilliesieae are mostly Chilean endemics; they are relatively rarely represented in herbaria, and are generally not available in cultivation (Zollner and Arriagada 1998; M. F. Fay unpubl.obs.).
Generic delimitation has been considered somewhat problematic in Gilliesioideae, and many species have been variously treated as members of several different genera.For example, /pheion unifiorum (Lindl.)Raf. has been included in Beauverdia Herter, Brodiaea Sm., Hookera Salisb., Leucocoryne Lindl., Milia Cav., Tristagma Poepp., and Triteleia Douglas ex Lindl., and Nothoscordum fragrans Kunth has been included in Allium, Geboscon Raf., Hesperocles Salisb., Maligia Raf., Milia, Ornithogalum L., and Sowerbaea Sm.Referring to Ipheion sellowianum (Kunth) Traub and I. dialystemon Guaglianone, Mathew ( 1995) stated that "these differ most noticeably from the blue I. unifiorum in having bright yellow flowers and, in fact, these 'ipheions' are probably better referred to as species of Nothoscordum."Both Ipheion and Nothoscordum were shown to be polyphyletic by Fay et al. (1997) using combined sequence data for rbcL and the trnL-F region.The delimitation of genera in the Ipheion/Nothoscordum complex remains one of the major unresolved issues in Alliaceae.In this paper we present further molecular studies of Alliaceae, with an emphasis on clarifying generic relationships within Gilliesioideae.Here we present an analysis of combined rbcL, trnL-F, rpsl6 intron, and ITS sequence data with increased taxon sampling compared to our earlier analyses.

MATERIALS AND METHODS
Species used for this study are similar to those in previous papers (Chase et al. 1995;Fay and Chase 1996;Meerow et al. 1999;Chase et al. 2000;Fay eta!. 2000).A full table of taxa including voucher information and GenBank accession numbers will be provided in a future paper.Methods of sequence production have varied greatly over time; primers and protocols can be found in studies of the individual loci (see Chase et al. [2006] for general comments, and Taberlet et al. [ 1991] for the trnL-F region, Oxelman et a!.[ 1997] for rpsl6, and Baldwin [1992] for ITS).The combined matrix included 23 taxa (21 members of Gilliesioideae, Tulbaghia, and Agapanthus).Agapanthus was defined as the outgroup based on the results of Fay and Chase (1996), Meerow et al. (1999), andFay et al. (2000).Allioideae were excluded from these analyses, due to problems with alignment of trnL-F and ITS sequences for this subfamily with Gilliesioideae, Tulbaghia, and Agapanthus (Fay et al. 2000 for trnL-F; this paper for ITS).
We analyzed the combined matrix using the branch and bound search option of PAUP* vers.4.0bl0 (Swofford 2001).We then used bootstrapping to estimate internal support with 1000 replicates of simple taxon addition, again using the branch and bound option.We show a randomly selected single tree to illustrate branch lengths (DELTRAN optimization due to problems with ACCTRAN optimization in PAUP* vers.4.0b10) and indicate groups not found in the strict consensus with arrows.We report all bootstrap support (BS) >50%.

RESULTS
The combined matrix consisted of 4480 characters of which 232 were excluded (mostly at the beginnings and ends of the four loci).Of the remaining 4148 characters, 791 were variable and 454 were potentially parsimony informative.The combined analysis gave eight equally parsimonious trees of Fitch length 1249 steps, consistency index 0.77, and retention index 0.80.One of the trees is shown in Fig. 1 (with bootstrap percentages), with arrows indicating branches that do not occur in the strict consensus tree.The trees were generally well resolved, with the only polytomy being found in one clade of four species of Nothoscordum.
Agapanthus was defined as the outgroup, and Tulbaghia (Tulbaghioideae) fell as sister to Gilliesioideae.Gilliesioideae were strongly supported as monophyletic (100% BS).Within Gilliesioideae, the genera with zygomorphic flowers (tribe Gilliesieae, represented here by Gilliesia and Solaria Phil.) formed a well-supported clade (100% BS).The genera with actinomorphic flowers (lpheion and allies) also formed a well-supported clade (100% BS), in which /pheion and Nothoscordum were both shown to be biphyletic.One clade of Ipheion (the species with yellow flowers) was intercalated between the two clades of Nothoscordum, and species of Leucocoryne s.l., Nothoscordum, and Tristagma were intercalated between the two clades of Ipheion.In both cases, the lack of monophyly was supported by high bootstrap percentages.Tristagma was shown to be monophyletic, with intermediate BS (79%).Leucocoryne s.l. was placed as sister to the blue/white flowered Ipheion spp.plus Tristagma, but this relationship gained only weak support (56%).At this level of sampling in Leucocoryne s.l., the taxa with six stamens (Pabellonia and Stemmatium) formed a clade sister to Leucocoryne s.s.

DISCUSSION
The regions of DNA used in this study were all easily alignable across the included taxa, but it was not possible to include Allium (Allioideae) due to a large insertion and a large deletion in the trnL-trnF intergenic spacer and markedly divergent ITS sequences in Allium relative to the other taxa in Alliaceae and Agapanthus.These features made se-quence alignment difficult or impossible.However, other analyses with length-conserved regions of DNA demonstrate the monophyly of Alliaceae, with Allioideae sister to Tulbaghioideae plus Gilliesioideae (e.g., Fay and Chase 1996).
Here we focus on relationships in Gilliesioideae.
Our results agree with those from other studies that demonstrate two clearly marked groups within Gilliesioideae.Sampling within Gilliesieae remains poor, as a result of the difficulty in obtaining material of these little-known plants.Thus we are unable to comment on an appropriate taxonomic treatment for the members of this tribe.Some genera are of uncertain affinity within the subfamily, and increased sampling is necessary to resolve these questions.Despite some obvious differences in floral structure (Rudall et al. 2002), the level of sequence divergence between Gilliesia and Solaria is low compared to that found in the Ipheion!Nothoscordum group; increased sampling might improve our understanding of the number of genera that should be recognized.This is currently somewhat variable.For example, Rahn (1998a) treated Ancrumia Harvey ex Baker and Gethyum Phil.as synonyms of Solaria, in contrast to Zollner and Arriagada (1998) and Rudall et al. (2002), who treated Ancrumia and Gethyum as separate genera.Inclusion of other South American genera of Alliaceae, such as Schickendanziella (Speg.)Speg., Speea Loes., and Trichlora Baker (listed as Gilliesioideae of uncertain placement by Rudall et al. 2002) would aid our understanding of this subfamily if material becomes available.
As stated by Fay eta!. (1997), neither Ipheion nor Nothoscordum as currently circumscribed is monophyletic, and this will result in some taxonomic rearrangements.All species with only one flower per umbel have been considered members of /pheion, but this analysis shows the emphasis on this single character to be unjustified.With this level of sampling, flower color among these uniflorous species seems to be more important, with the blue-flowered species falling with Tristagma and the yellow-flowered species falling with one group of Nothoscordum, some of which are also yellow flowered.However, the relatively low taxon sampling combined with the lack of strong support for one critical branch defining relationships between the genera, namely the branch indicating a relationship between the blue-flowered /pheion plus Tristagma with Leucocoryne s.l., means that taxonomic rearrangement may be premature.The taxonomic history of these taxa is already complex, and adding a further set of names at this stage would make it more so.Inclusion of other taxa (e.g., the genera treated as synonyms by Rahn (1998a) including Zoellnerallium Crosa [= Nothoscordum], and Garaventia Looser and Steinmannia Phil.[ = Tristagma]) would be desirable.Taxonomic changes will be the subject of a future paper.
Lack of unambiguous morphological characters supporting the subgroups within the clade composed of lpheion, Nothoscordum, and allies is also a problem.Single-flowered inflorescences have evolved at least twice, and reduction in stamen number appears to be a unique character in Leucocoryne s.s.Other species of Leucocoryne s.l.(e.g., L. narcissoides Phil.= Stemmatium narcissoides Phil.) possess six stamens.Stamen number also shows reduction in some members of Gilliesieae, but this is related to floral zygomorphy encompassing both stamen whorls, and is indepen-dent of the actinomorphic reduction involving a single stamen whorl in Leucocoryne (e.g., Zollner and Arriagada 1998;Rudall et al. 2002).Some of the other characters that have been used to separate the genera in the clade composed of Ipheion, Nothoscordum, and allies include continuously variable characters such as the degree of fusion of tepals or spathe bracts; our analysis indicates that these are also ambiguous.Other characters that merit further investigation include seed coat texture, reported as pitted in lpheion and smooth in the other genera (Rahn 1998a).
Fig.1.-0ne of eight most-parsimonious trees from the combined analysis, tree length 1249 steps, consistency index 0.77, retention index 0.80.Branch lengths are given above the lines, bootstrap percentages >50% below.Arrows indicate the two branches that do not occur in the strict consensus tree.