Phylogenetics of Lilliales

In order to investigate interfamilial relationships of Liliales we analyzed a combined matrix of plastid rbcL, trnL intron, trnL-F intergenic spacer, rnatK, and ndhF, and mitochondrial atpl DNA sequences. The results are generally congruent with previous broad analyses and provide higher bootstrap support for many relationships. Important changes relative to previous studies are the recognition of Petermanniaceae distinct from Colchicaceae and the tentative inclusion of Corsiaceae in the order. This brings the number of families in the order from nine to eleven. The additional data presented here strengthen the case for including Uvulariaceae in Colchicaceae and Calochortaceae in Liliaceae.


INTRODUCTION
In recent classifications (Chase et a!. 2000;Angiosperm Phylogeny Group II [APG II 2003]), the order Liliales consists of nine families: Alstroemeriaceae, Campynemataceae, Colchicaceae, Liliaceae, Luzuriagaceae, Melanthiaceae, Philesiaceae, Rhipogonaceae, and Smilacaceae.This ordinal circumscription is generally similar to that of Dahlgren, Clifford, and Yeo (1985) but with some marked contrasts, notably the exclusion of Iridaceae and Orchidaceae (both in Asparagales in Chase et al. 2000;APG II 2003).The background to and circumscription of the order Liliales are discussed in detail elsewhere (Fay and Chase 2000;Rudall et al. 2000).Colchicaceae (including Uvulariaceae), Liliaceae (including Calochortaceae sensu Tamura [ 1998)]; see Discussion) and Melanthiaceae (including Trilliaceae and Xerophyllaceae, but excluding Nartheciaceae, Petrosaviaceae, and Tofieldiaceae) are at variance with previous classifications, e.g., Dahlgren et al. (1985).Genera included in the II families of Liliales as circumscribed here are listed in Table 1, as well as their different placements in earlier systems.Two additions at the family level to Liliales are made: Corsiaceae and Petermanniaceae (see below).
In this paper, we discuss the interrelationships of these 11 families on the basis of combined analyses of plastid trnL intron and trnL-F intergenic spacer (together known as the trnL-F region), rbcL, ndhF, and matK, and the mitochondrial atp 1.In addition, we focus in more detail on relationships within Liliaceae, following on from the studies of Fay and Chase (2000) and R0nsted et al. (2005).

MATERIALS AND METHODS
Species used as placeholders for this study are similar to those in previous papers (Chase et al. 1995(Chase et al. , 2000;;Rudall et al. 2000; R!Z)nsted et a!.2005).For newly produced data (since Chase et a!. 2000 andRudall et al. 2000), we exchanged DNA samples (notably true Petermannia) between the participating laboratories so that each locus was amplified from the same genomic DNA in most cases.Species used are listed in Table 2. Methods of sequence production have varied greatly over time; primers and protocols can be found in studies of the individual loci (summarized in Chase et a!. 2006, with the addition of Taber let et al. (1991) for the trnL-F region).
The combined matrix contained 36 taxa, including Pandanus L. f. and Stemona Lour.as outgroups based on the results of Chase et al. (2000Chase et al. ( , 2006)).We analyzed the combined matrix using heuristic searches with PAUP* vers.4.0b10 (Swofford 2002) using the following strategy: 500 replicates of randomized taxon entry with subtree-pruningregrafting (SPR) swapping and a tree limit of 20 trees per replicate to reduce the time spent swapping on suboptimal islands of trees.In a second round of analysis we used these as starting trees with tree-bisection-reconnection (TBR) swapping to find any additional trees.We then used bootstrapping to estimate internal support with 500 replicates of simple taxon addition, again with a limit of 20 trees per replicate.DELTRAN optimization is used to illustrate branch lengths, due to problems with ACCTRAN optimization in PAUP* vers.4.0b10.We report all bootstrap percentages (% BS) >50.The analyses were repeated excluding Arachnitis and Petermannia to evaluate potential problems due to missing data.

RESULTS
The aligned matrix contained 9141 characters, of which 1128 were excluded (mostly in the trnL-F region due to   Vinnersten and Reeves (2003) showed that several genera of Colchicaceae as currently circumscribed are not monophyletic.Notably, Bulbocodium, Colchicum, and Merendera are embedded in Androcymbium, whereas Gloriosa and Littoria are intermingled, and Onixotis is possibly paraphyletic to Wurmbea.Thus, a taxonomic revision of the family at the generic level is necessary.' Dahlgren et al. (1985) did not mention Burchardia, and it is not possible to place it in their system.
d n/a indicates that the genus was described after this taxonomic treatment.' Prosartes was considered in most previous systems to be a synonym of Disporum (Colchicaceae).However, North American species are members of Liliaceae, whereas Asian species are members of Colchicaceae.Prosartes applies to the North American species.
'Genera as recognized by Zomlefer et al. 2005.g Previously treated as Zigadenus p.p.   1], and excluding these taxa, results not shown) gave the same topologies for the families of Liliales included in common, with only minor variation in bootstrap support on some branches.Here we present the data from the analysis with all taxa included.The analysis resulted in two trees (tree length 5736 steps, consistency index 0.60, retention index 0.60).Liliales were strongly supported (100% BS).The only difference between the trees pertained to the relative positions of Campynemataceae (here represented by Campynema) and Arachnitis as the first branch/es in the order, with Arachnitis and Campynema: (1) as successive sisters to, or (2) together as sister to the rest of the order.We show result ( 1 ), marking the branch that collapses in the strict consensus with an asterisk (Fig. 1).
The rest of the order (exclusive of Arachnitis and Campynema) was moderately supported (70% BS) as monophyletic.Melanthiaceae were then sister to the remaining families of Liliales, but with bootstrap support <50%.The other families fell in two clades.The first clade, with weak support (64% BS), was (Petermanniaceae (Colchicaceae (Alstroemeriaceae + Luzuriagaceae))).However, the branches defining the relationships between the four families within this clade all gained strong support (98-100% BS).The second, with strong support (99% BS), was ((Philesiaceae + Rhipogonaceae) (Smilacaceae + Liliaceae)).Bootstrap support was strong for the branches defining the sister group relationship between Smilacaceae and Liliaceae (94% BS) and the monophyly of both Liliaceae and Philesiaceae (both 100% BS).The sister group relationship between Philesiaceae and Rhipogonaceae was only weakly supported (54% BS).

Relationships of Liliales
The data presented here do not allow us to address the issue of the placement of Liliales with respect to other monocot orders.However, in the analyses of Chase et al. (2006)  s is, 68 % BS for the all-locus ana lys is).Chase et al. (2006) recommend the use of additi onal data and analysis to resolve thjs critical node in the monocots.If their topology ho lds in further a na lyses, Lilianae sensu Dahl gre n et a l. ( 1985) are paraphyletic.

Circumscription of Liliales
The o nl y sign ificant c hange to the ci rc umscripti on of the order, relative to RudaLI et a l. ( 2000) and Chase et al. (2000) is the te ntati ve inc lu sion of Corsiaceae, o n the basis of the position of Arachnitis in the analyses of Chase et a!. (2006), based on sequences for nuclear ribosomal ISS and mitochondrial atp I. Neyland and Hennigan (2003), using partial sequences for 26S nuclear ribosomal DNA alone, suggested that Corsiaceae may be polyphyletic.In their analysis, Corsia fell with Liliales, whereas Arachnitis fell with Thismia Griff.(Dioscoreales).Arachnitis and Thismia are both achlorophyllous, and plastid data for the former are lacking in the analyses of Chase et a!. ( 2006), but we consider the result to be more robust than that of Neyland and Hennigan (2003) as it is based on data from nuclear and mitochondrial genomes and on a larger number of data points (base pairs).Rudall and Eastman (2002) examined the relationships of Corsia on the basis of floral anatomy and pollen morphology and found evidence that could support a relationship to Campynemataceae or Thismia.Thus, it appears that the placement of Corsiaceae remains problematic, and a definitive placement and assessment of its monophyly must await further data.
Petermannia falls as sister to a group of three families (Alstroemeriaceae, Colchicaceae, and Luzuriagaceae) rather than within Colchicaceae as in previous analyses.Thus, Petermanniaceae should be resurrected.This change in position is because the material labelled as Petermannia in earlier analyses was later shown to be misidentified Tripladenia cunninghamii.For further information, see Chase et a!. (2006).

Family Relationships
Interrelationships among the families are the same (with the exception of Smilacaceae, see below) as that obtained by Chase et a!. ( 2006), despite differences in taxon sampling and the loci used.Bootstrap support is also generally similar, but notable differences are the weak support for the position of Rhipogonaceae as sister to Philesiaceae [54% BS; cf.Chase et a!.2006, 100%] and the strong support for the position of Smilacaceae as sister to Liliaceae [94%;cf. Chase et a!. (2006), where Smilacaceae fell as sister to (Philesiaceae + Rhipogonaceae) + Liliaceae, with only 56% BS].The low level of support for the pattern of relationships between these families in the analysis of Chase et a!. ( 2006) may be a reflection of the relatively low sampling density or the different loci used (in their plastid-only analysis, Smilacaceae fell as sister to Liliaceae, but again with weak BS).

Family Circumscriptions
Melanthiaceae are here recognized in a broad sense, including Trilliaceae, and this treatment is supported strongly by bootstrap analysis here and in the analyses of Zomlefer et a!. (2006).Their analyses should be referred to for intrafamilial relationships, as these include far more taxa in the family than do those presented here (we only used placeholders).
In Colchicaceae, Uvularia is strongly supported as sister to the rest of the family as sampled here, and the family is also strongly supported.Vinnersten and Reeves (2003) analyzed relationships within Colchicaceae using three plastid DNA regions and much wider taxon sampling.The relationships found here are in agreement with their analyses.They included Burchardia (not included in our study) in their analyses, and this fell further outside core Colchicaceae than Uvularia.However, in an earlier study with less sampling and only using rbcL, the positions of Uvularia and Burchardia were reversed (Vinnersten and Bremer 2001).Also, Burchardia was included in Colchicaceae by APG (1998), APG II (2003), and Vinnersten and Reeves (2003).Due to the lability of the relative positions of Burchardia and Uvularia and in view of these recent taxonomic treatments, we recommend that Uvularia be included in Colchicaceae and that Uvulariaceae should not be recognized.Several genera in Colchicaceae are not monophyletic in the study of Vinnersten and Reeves (2003; see also footnote to Table 1).
Luzuriaga (Luzuriagaceae) is strongly supported as sister to Alstroemeriaceae, and these families could be combined.However, we choose not to do this for the time being in order to maintain the stability of family circumscriptions.Both Alstroemeriaceae and Luzuriagaceae have been recognized in most recent treatments (e.g., Chase et a!. 2000;APG II 2003).The two families do, however, have shared characteristics, including the possession of inverted leaf blades (Dahlgren et a!. 1985).
In Liliaceae, the positions of Calochortus and Tricyrtis are still not well defined.In both trees, Calochortus is sister to the clade containing Prosartes, Scoliopus, and Streptopus, but with BS <50%.Tricyrtis is sister to the remaining genera ("core Liliaceae" = Lilioideae and Medeoloideae), but again with BS <50%.This pattern of relationships mirrors that of R0nsted et a!. (2005).In their analyses of ITS, matK and the rpll6 intron, with considerably greater sampling in Liliaceae, Tricyrtis was sister to core Liliaceae with 91% BS.The position of Calochortus was, however, only weakly supported (67% BS).In the earlier study of Fay and Chase (2000), Tricyrtis was placed as sister to the Prosartes/Scoliopus/Streptopus clade and the placement of Calochortus was unresolved.Thus, the phylogenetic relationships of Calochortus and Tricyrtis have proved problematic.The relationships obtained here and by R0nsted et a!. (2005) render Calochortaceae sensu Tamura ( 1998) paraphy letic.Patterson and Givnish (2003) used both Calochortaceae and Liliaceae in one article for the family containing Calochortus, and there is clearly a need for stability in family circumscription.For these reasons, we choose to recognize a wide circumscription of Liliaceae (as in Chase et a!. 2000, andAPG II 2003).
In Lilioideae, the position of Gagea as sister to Tulipa + Amana was only weakly supported.However, with increased taxon sampling, R0nsted et a!. ( 2005) recovered the clade ((Gagea + Lloydia) (Tulipa (Amana + Erythronium))) with 97% BS.They used the tribal name Tulipeae for this clade.This varies from previous studies (see Rudall et a!. [2000] and Fay and Chase [2000] in which Tulipa, Amana, and Erythronium were placed as sister to the remainder of core Liliaceae, including Medeoloideae, although these patterns lacked strong bootstrap support).Improving taxon sampling and increasing amounts of data have allowed us to achieve greater resolution and support within Liliaceae.Our data and those of R0nsted et a!. ( 2005) both indicate a sister relationship of Tulipeae to Lilieae ( Cardiocrinum, Fritillaria, Lilium, Notholirion), but only with low to moderate support.
Fritillaria and Lilium again are shown to be closely related.In our earlier studies with fewer loci (summarized in Fay and Chase 2000), we found two clades of Fritillaria and one of Lilium (including Nomocharis Franch.), but the relationship between the three clades was not well resolved.Thus, we were not able to demonstrate the monophyly of Fritillaria.The analysis presented here does not allow us to address this question (as we only include one species per genus), but the study conducted by R¢nsted et al. (2005), in which multiple species of both genera were included, supports the hypothesis that both genera are monophyletic.Cardiocrinum and Notholirion are successive sister groups to Fritillaria + Lilium in that study, but the branch separating Notholirion from (Cardiocrinum (Fritillaria + Lilium)) is only weakly supported.

Prospects for Improvement
The main remaining problem in Liliales relates to the inclusion of Corsiaceae.Given their achlorophyllous nature and the extreme rarity of Corsiopsis, it may not prove possible to obtain plastid DNA data for these taxa, and it is unlikely that improving taxon sampling will be easily achieved, despite its obvious desirability.Collection of further nuclear and mitochondrial sequences for the taxa for which DNA is already available appears to be the best way forward in addressing this problem.
Elsewhere, there is scope for improving taxon sample or increasing the number of loci to address particular nodes where support is weak.The most significant of these in terms of interfamilial relationships relate to the position of Campynemataceae (and Arachnitis) as sister to the rest of the order and the relative positions of Melanthiaceae and the remaining two large clades that form a trichotomy in the bootstrap tree.Within Liliaceae, the relationships of Calochortus (and, to a lesser extent, Tricyrtis, Notholirion, and Cardiocrinum) still require further investigation.

Fig
Fig. I.-One of the two most-parsimonious trees obtained (DELTRAN optimi zation).Tree le ngth = 5736 steps, consistency index = 0.60, retenti on index = 0.69.Bra nch le ngth s are g ive n above th e branches, bootstrap perce ntages below.The branch that coll apses in the strict co nsensus is indicated w ith an asterisk.Bars and letters by groups within Li liaceae relate to the c lades referred to in the Results section.

Table I .
The families and genera of Liliales as recognized in this study.Where placements differ from four previous classifications, the alternative placement is given.The placement of Corsiaceae in the order is tentative at this stage.Genera included in this study are marked with an asterisk [*].

Table l .
Continued.Placements are given in the index of Wilson and Morrison 2000.The families recognized are those listed in Chase et al. 2000.

Table 2 .
Taxa included in this study with voucher information.All loci were sequenced for the same species, except as indicated in the final column.If no locus is given in the final column, then this species was the default.