A phylogenetic evaluation of a biosystematic framework: Brodiaea and related petaloid monocots (Themidaceae)†
The authors thank the following individuals for providing field assistance or plant materials: Lowell Ahart, Gery Allen, Steve Boyd, Mary Carlson, Mark Chase, Curtis Clark, Holly Forbes (University of California Botanical Garden), Craig Freeman, Judy Gibson, Dylan Hannon, Russell Holmes, Thad Howard, Larry Hufford, Kyle Ince, Steve Jessup, Glenn Keator, Marla Knight, Lee Lenz, Aaron Liston, Daniel Luoma, Kim Marsden, Dale McNeal, Julie Nelson, Vern Oswald, Jon Rebman, Suzanne Rhodes, Aaron Rodriguez, Andy Sanders, Garey Schoolcraft, Charlene Simpson, Mike Simpson, Veva Stansell, Richard Straw, and Kirsten Winter. The authors thank Kandis Elliott and Claudia Lipke for producing the figures, the University of Wisconsin Biotechnology Center for facilitating automated sequencing, and Cora Fox, Sean Graham, and Dale McNeal for reviewing the manuscript. This project was supported by the following institutions: the National Science Foundation (DEB 99012169 to JCP and KJS, and DEB 9306943), the Hunt Institute of Carnegie-Mellon University, the Botanical Society of America, the American Society of Plant Taxonomists, the California Native Plant Society, the Oregon Native Plant Society, Sigma Xi, and the Davis, Raper, O. N. Allen, and Vilas Funds at the University of Wisconsin.
Abstract
Phylogenetic analyses of plastid DNA sequences of ndhF, trnL-F intron and spacer regions, and rpl16 are presented separately and combined for 41 taxa from all 12 genera of the Themidaceae and for 20 taxa from nine related families in the higher Asparagales. The results from the combined analysis are the most resolved and provide a high level of support for the monophyly of Themidaceae. Within Themidaceae, the Milla complex of Mexico is supported as monophyletic within a paraphyletic Brodiaea complex of western North America. Four major clades are identified in each of the individual and combined analyses: (1) the Milla complex; (2) Brodiaea, Dichelostemma, and Triteleiopsis; (3) Triteleia, Bloomeria, and Muilla clevelandii; and (4) Androstephium and the other species of Muilla. These well-defined clades suggest that morphological characters (e.g., an extended perianth tube) that have been traditionally used to circumscribe the genera within the Brodiaea complex have evolved independently at least twice. In addition, common biogeographic distribution patterns (e.g., Brodiaea and Triteleia having centers of diversity in northern California and the Pacific Northwest) appear to be the result of separate evolutionary radiations.
The resurrected plant family Themidaceae Salisb. contains 12 genera of perennial petaloid geophytes found principally in western North America (Fay and Chase, 1996; Pires et al., 2001). These genera, formerly recognized as tribe Brodiaeae in the Alliaceae (e.g., Dahlgren, Clifford, and Yeo, 1985), have been previously divided into two complexes: the enigmatic Milla complex, centered in Mexico, and the well-studied Brodiaea complex, centered in the western United States. The Mexican genera (e.g., Bessera Schult, Dandya H. E. Moore, Milla Cav., and Petronymphe H. E. Moore) have brilliant flowers that attract pollinators such as hawk moths and hummingbirds, but the complex is incompletely known, with several new species and a new genus having been described in the past 10 yr (Delgadillo, 1992; Lopez-Ferrari and Espejo, 1992; Turner, 1993; Howard, 1999). Similarly, some of the satellite genera of the Brodiaea complex (e.g., Androstephium Torrey, Bloomeria Kellogg, Muilla S. Watson ex Benth., and Triteleiopsis R. F. Hoover) have been left as taxonomic mysteries (Ingraham, 1953; Shevock, 1984; White, Sanders, and Wilcox, 1996). However, Brodiaea sensu lato (s.l.) (Brodiaea Smith, Dichelostemma Kunth, Triteleia Douglas ex Lindl.) has been a classic biosystematic example of evolutionary radiation with rampant polyploidy and a kaleidoscopic range of floral diversity with variation in flower position and perianth color and architecture, as well as various elaborations of the androecium (Moore, 1953; Keator, 1967, 1989; Niehaus, 1971, 1980). Brodiaea s.l. displays a striking degree of ecological endemism, with species often restricted to serpentine outcrops, vernal pools, or marine terraces on islands off the coasts of southern California and Mexico. Raven and Axelrod (1978) considered Brodiaea one of the most important genera for understanding the evolution of the Californian flora.
The genus Brodiaea Smith has a rich taxonomic history that includes many controversies (Hoover, 1939a; Keator, 1967, 1989; Niehaus, 1971, 1980). One of the primary debates has been over the presence of Brodiaea in South America (Table 1). While Baker (1896) included 45 species of geophytes native to both North and South America in the genus, Hoover (1939a) defined Brodiaea as strictly North American. Following Watson (1879) and Greene (1886, 1890), Hoover separated the North American species from the South American species using the presence of several characters: jointed pedicels, fibrous-coated corms, several bracts subtending the inflorescence, a tubular perianth with distinct filaments, and the lack of an alliaceous (onion-like) odor. Niehaus (1980) endorsed Traub's (1963) contention that the South American “brodiaeas” should be placed in the genera Tristagma and Nothoscordum in subtribe Alliinae. In addition, Niehaus (1980) agreed with Traub's (1963) placement of Brodiaea and allies in the family Amaryllidaceae, tribe Allieae, subtribe Brodiaeinae. Biosystematic researchers (Niehaus, 1980; Keator, 1989) have continued to amplify the many differences between the North and South American “brodiaeas” (Table 1).
A second controversy among California biosystematists was whether Brodiaea should be considered one or three genera (Table 2; see Keator, 1967, 1989; Niehaus, 1971, 1980). Hoover (1939a) agreed with Greene (1886) and Jepson (1925) that the Brodiaea of North America could be further distinguished into three groups. While Jepson regarded these three distinct groups as subgenera of Brodiaea, Hoover considered them as three separate genera: Brodiaea, Dichelostemma Kunth, and Triteleia Douglas ex Lindl. Keator (1967, 1989, 1993a–c), and Niehaus (1968, 1971, 1980) supported Hoover's original distinction of three genera, citing additional morphological characters, chromosome counts, and hybridization studies (Table 2). Thus, Hoover's (1939a, b) definition of North American Brodiaea s.l. as three related genera has remained unchallenged, and recent studies and taxonomic treatments have continued to support the idea that Brodiaea should be recognized as a small genus of ca. 14–15 taxa related to Dichelostemma (five taxa) and Triteleia (ca. 15 taxa) on the basis of an extended perianth tube (Niehaus, 1980; Keator, 1993a–c) and a shared geographic center along the Pacific Coast of western North America (Burbank, 1941). Given even this simplified taxonomic history of Brodiaea s.l., it is no surprise that modern biosystematic studies and current taxonomic treatments have considered Brodiaea, Dichelostemma, and Triteleia to be closely related. However, while the taxonomy of Brodiaea s.l. has been well deliberated for over a century (e.g., Wood, 1868; Baker, 1871; Watson, 1879; Greene, 1886; MacBride, 1918; Jepson, 1925; Hoover, 1939a; Keator, 1967, 1989; Niehaus, 1971, 1980), the other genera of Traub's subtribe Brodiaeinae have not received as much attention from taxonomists (Traub, 1963).
The taxonomic controversies regarding Brodiaea and related taxa have resulted in four notable levels of evolutionary relationships that have been assumed by biosystematists for over five decades (between 1939 and 1993): (1) Brodiaea and relatives in subtribe Brodiaeinae are related to Alliaceae on the basis of sharing a scapose umbelliferous inflorescence and superior ovary and are thus separate from Liliaceae (superior ovary but no umbels) and Amaryllidaceae (umbels but inferior ovary); (2) Brodiaea and relatives in subtribe Brodiaeinae are united by the presence of a corm for rootstock in contrast to a tunicated bulb found in Alliaceae; (3) the Brodiaea complex and Milla complex are separated by corm coat characters and geographic centers of distribution; and (4) Brodiaea s.l. (Brodiaea, Dichelostemma, and Triteleia) is considered a separate taxonomic entity from the other genera in the Brodiaea complex on the basis of a shared extended perianth tube and more northerly center of biogeographic distribution (Fig. 1). Therefore, much of the biosystematic literature between 1939 and 1993 assumed that Brodiaea, Dichelostemma, and Triteleia were related and compared them morphologically, cytologically, and ecologically with each other (Table 2) and with the alliaceous South American “brodiaeas” (Table 1), while generally ignoring the relationships of Brodiaea s.l. to the satellite genera of the Brodiaea complex and to the genera of the Milla complex.
Recent studies have examined all four of these evolutionary assumptions using phylogenetic methods and a combination of morphological and molecular characters. First, the assumption that Brodiaea and relatives are closely related to Allium and other alliaceous taxa (including South American “brodiaeas”) is not supported; rather, Brodiaea and relatives appear to have closer affinities to Hyacinthaceae and other families in the higher asparagoids of the Asparagales (Fay and Chase, 1996; Fay et al., 2000; Pires et al., 2001). This exclusion of the tribe Brodiaeeae from Alliaceae is supported by DNA sequence data and several morphological and anatomical characters (Fay and Chase, 1996; Pires et al., 2001). These same studies validated the second assumption: that the tribe Brodiaeeae forms a group of related genera. For these reasons, Fay and Chase (1996) resurrected Themidaceae Salisb. for Brodiaeeae, based on Themis ixioides (Ait. f.) Salisb. [= Triteleia ixioides (Ait. f.) Greene] (Salisbury, 1866), and recent data continue to strongly support the monophyly of this recircumscribed Themidaceae (Meerow et al., 1999; Chase et al., 2000; Fay et al., 2000; Pires et al., 2001).
Pires et al. (2001) began to address the other two phylogenetic assumptions by analyzing 25 morphological characters and sequences of rbcL and the trnL-F region of plastid DNA. The third assumption—that the Milla complex and Brodiaea complex, as defined by Moore (1953), are monophyletic assemblages—was supported by this recent study. However, the support was moderate, and the family's circumscription remains unclear. Turner (1993), for example, erected and placed a monotypic Jaimehintonia Turner in the Brodiaea complex even though Jaimehintonia occurs in the same range as the Milla complex of Mexico. Pires et al. (2001) tested the fourth assumption, the monophyly of Brodiaea s.l. (Brodiaea, Dichelostemma, and Triteleia), and found evidence that this assumption was problematic: in a combined analysis of molecular and morphological data, Triteleia was strongly supported to be sister to Bloomeria at the level of taxon sampling in that study, and those two genera were weakly supported as sister to Androstephium and Muilla. Furthermore, Brodiaea was found to be embedded in Dichelostemma, although only 2 of the 15 species of Brodiaea were sampled. Collectively, these recent studies have begun to provide a more complete picture of the relationships between the families of the asparagoid monocots and between the genera of the Themidaceae. However, many questions remain concerning the evolutionary relationships between lineages within the higher asparagoids of the Asparagales, between genera of the Themidaceae, and between species within the genera.
In the study reported here, we reexamined the four levels of assumed phylogenetic relationships described above (Fig. 1) with a two-tiered taxonomic and genetic sampling. Initially, we examined the relationships between Themidaceae and nine families in the higher Asparagales using plastid DNA sequences of ndhF and trnL-F to verify the phylogenetic disjunction between Themidaceae and Alliaceae and to define an appropriate outgroup for the second level of sampling. Although trnL-F has been previously used to evaluate these relationships (Meerow et al., 1999; Fay et al., 2000), our study is the first to use ndhF to evaluate higher level relationships in the Asparagales. For the remainder of the study, we evaluated relationships within the Themidaceae by using plastid DNA sequences of ndhF, trnL-F, and rpl16 and extending previous studies with a more complete taxonomic sampling of Themidaceae, but with a reduced number of outgroups. By adding three previously unsampled genera (Triteleiopsis, Petronymphe, and Jaimehintonia), we conducted the first study to assess all 12 genera postulated to belong to the Themidaceae. Here, we endeavor to: (1) Confirm whether Themidaceae is sister to Alliaceae or to other families in the higher Asparagales and thus whether an umbelliferous scape has arisen one or more times; (2) Test the monophyly of the western North American cormous Themidaceae with a taxon sampling that includes all the genera; (3) Reevaluate the monophyly of Moore's two complexes; and (4) Establish the relationships between the genera of the Themidaceae. The phylogenetic analyses from the three molecular data sets are significant because they suggest that key morphological characters (e.g., the extended perianth tube used to define Brodiaea s.l.) and biogeographic distribution patterns (e.g., Pacific Coast center of the Brodiaea complex) that have been traditionally used to circumscribe the genera and complexes within the Themidaceae are problematic.
MATERIALS AND METHODS
Taxon sampling
All 12 genera of the Themidaceae were sampled, including 41 of the 62 currently recognized species. Twenty taxa from nine related families in the higher Asparagales were selected as outgroups (following Fay and Chase, 1996; Meerow et al., 1999; Chase et al., 2000; Fay et al., 2000; Pires et al., 2001). Sampling density within Themidaceae was extensive (see Appendix [http://ajsupp.botany.org/v89/]). For seven genera, all recognized species were sampled (Androstephium, Bloomeria, Dichelostemma, Jaimehintonia, Muilla, Petronymphe, and Triteleiopsis). Three genera were also well sampled: two of three species of Bessera, 9 of 14 species of Brodiaea (with exemplars from all five recognized sections within Brodiaea), and 10 of ca. 15 species of Triteleia (with exemplars from all three recognized sections within Triteleia). Only one of four species of Dandya was sampled owing to lack of material, and two of six species of Milla were sampled. Most Themidaceae were collected during field trips from 1994 to 1999, while representatives of outgroup taxa were usually provided by botanical gardens or other researchers.
From each of these taxa, two to three cpDNA regions were sampled (see Appendix [http://ajsupp.botany.org/v89/]), depending on each taxon's role in two sampling strategies. The two strategies were designed to resolve relationships (1) among families in the higher Asparagales and (2) within the Themidaceae. Regions of the cpDNA with the lowest divergence (ndhF and trnL-F) were used to resolve the oldest cladogenetic events among families, while an additional region with the greatest divergence (rpl16 intron) was also used to help resolve more recent ones.
In the first analysis, we sampled the greatest taxonomic breadth with somewhat reduced taxonomic density in the Themidaceae. The sampling included (1) eight taxa of the Themidaceae, (2) eight taxa of the Hyacinthaceae (the putative sister family to Themidaceae per Fay and Chase, 1996), (3) four species of Agavaceae s.l., (4) two genera in the Alliaceae, and (5) one genus each in the Anthericaceae, Asparagaceae, Convallariaceae, Agapanthaceae, Amaryllidaceae, and Asphodelaceae (all families defined per Angiosperm Phylogeny Group, 1998). In the second sets of analyses, we used all the Themidaceae taxa mentioned above (see also Appendix), and for outgroups we used seven species of Hyacinthaceae and one species each of Agavaceae and Agapanthaceae based on the results of the first analysis and the ability to generate rpl16 sequence for those outgroup taxa.
DNA extraction, amplification, and sequencing
Fresh leaf tissue, leaf tissue dried with silica gel (Chase and Hills, 1991), or tissue from herbarium specimens was used. DNA was extracted from 1.0 g fresh, 0.2–0.25 g silica-dried, or 0.1–0.15 g herbarium material using a 6% (mass/volume) hexadecyltrimethyl ammonium bromide (CTAB) method with ethanol and ammonium acetate precipitations (Doyle and Doyle, 1987, with minor modifications as detailed in Smith et al., 1991).
The polymerase chain reaction (PCR; Mullis and Faloona, 1987) was used to amplify each cpDNA region. Each PCR reaction mix consisted of 31 μL double-distilled H20, 5 μL MgCl2, 5 μL buffer, 4 μL dNTP, 2.5 μL BSA (bovine serum albumin), 1 μL of each of two primers, and 0.5 μL enzyme. All PCR and cycle sequencing reactions were run on a Gene Amp PCR System 2400 (Perkin Elmer, Foster City, California, USA); specific PCR and sequencing primers for the three cpDNA regions are indicated below. To successfully detect amplified DNA and the possible contamination of negative controls, PCR products were examined on 1% agarose gels using ethidium bromide and ultraviolet light. Amplified products were purified with the spin columns in the QIAquick PCR Purification Kit (Qiagen, Valencia, California, USA) following protocols provided by the manufacturer. Sequencing product was precipitated in ethanol and sodium acetate to remove excess dye terminators before being run out on an ABI Prism 377 DNA Sequencer (according to the manufacturer's protocols; PE Applied Biosystems, Foster City, California, USA). Aligned tracefiles in Sequencer 3.0 (Gene Codes, Ann Arbor, Michigan, USA) were compared to detect mistakes and correct uncertainties in the computer-generated sequence.
Amplifying the entire ndhF region (primer 032F–2110R) in one PCR reaction was inefficient, so PCR was accomplished by amplifying two overlapping regions. The ndhF primers used for PCR and sequencing were designed for Vriesea (Bromeliaceae; Terry, Brown, and Olmstead, 1997). The front 5′ half of ndhF was amplified with primers 032F and 1318R (or, failing that, 032F–1101R, but the latter leaves a gap). The more variable 3′ end was amplified with primers 1101F and 2110R (or, failing that, 1318F and 2110R). The PCR conditions for ndhF were optimized at 94°C for 5 min, then 27 cycles of 94°C for 30 s, 52°C for 1 min, 72°C for 1.5 min, and finally 72°C for 7 min, followed by 4°C indefinitely. If there was insufficient PCR product, lowering the annealing temperature down to 48°C for 30 cycles routinely solved the problem. Four sequencing primers were used for complete coverage of the 5′ end of ndhF and an additional four primers for the 3′ end, so for each taxa eight sequencing primers were used to accomplish generous double overlap. The 5′ half sequencing primers were ndhF 32F, 451F, 745R, and 1101R. Since sequencing 1318R routinely failed, the 451F primer was critical because it crossed 1101 to link the 5′ and 3′ halves. The 3′ half sequencing primers were ndhF 1101F, 1318F, 1600R, and 2110R.
Amplification of the trnL-F region (which includes three subregions: the trnL intron, the 3′ exon of trnL, and the trnL-trnF intergenic spacer) was carried out using primers c and f of Taberlet et al. (1991). The PCR conditions for trnL-F were optimized at 94°C for 5 min, then 30 cycles of 94°C for 30 s, 48°C for 1 min, 72°C for 1.5 min, and finally 72°C for 7 min, followed by 4°C indefinitely. Primers c, e, and f were used for cycle sequencing.
Amplification of the rpl16 intron was accomplished by using primers F71 of Jordan, Courtney, and Neigel (1996) and R1661 (Kelchner and Clark, 1997). The PCR conditions for rpl16 were optimized at 94°C for 5 min, then 25 cycles of 94°C for 1.5 min, 54°C for 2 min, 72°C for 2 min, and finally 72°C for 7 min, followed by 4°C indefinitely. Primers F71 (Jordan, Courtne, and Neigel, 1996) and R1516 (Baum, Small, and Wendel, 1998) were used for cycle sequencing.
Sequence alignment: defining substitution and indel characters
Sequences of ndhF were easily aligned manually because very little length variation was detected. For trnL-F and rpl16, sequences of several taxa representing the range of probable variation in the matrix were aligned using the Clustal option in Sequence Navigator (PE Applied Biosystems, Foster City, California, USA), followed by manual optimization and alignment of subsequent sequences (Kelchner, 2000). The trnL-F region and rpl16 intron were easily aligned within the Themidaceae, and most of the alignments were straightforward even across the higher Asparagales. As a result, no DNA regions were ambiguous enough to warrant exclusion from the analysis.
Indels were coded as additional, unordered characters for maximum parsimony analyses if they were bordered by stretches of unambiguously aligned nucleotide base pairs and were potentially informative. Single base-pair insertions/deletions (indels) were not coded if they were adjacent to strings of the same nucleotide base pairs (e.g., four A's present vs. five A's). Other researchers have similarly excluded this type of indel because it may arise from methodological error (Downie et al., 1998; McDade and Moody, 1999) or because they have noted its evolutionary lability (Small et al., 1998). The indels were coded as the same state if they were the same size and did not vary by more than one nucleotide substitution (Baum, Sytsma, and Hoch, 1994; Graham et al., 2000; Simmons and Ochoterena, 2000).
Phylogenetic analysis
Data analysis was performed with PAUP* 4.0b4a (Swofford, 2000) using assumptions of maximum parsimony (with and without coding indels) and maximum likelihood, but because all preliminary analyses gave trees with similar topologies, only maximum parsimony analyses that included indels are reported here. Congruence between the cpDNA regions was evaluated using the incongruence length difference (ILD) test (= partition homogeneity test) (Mickevich and Farris, 1981; Farris et al., 1995; Swofford, 2000). The pairwise congruence of the relevant data sets was tested using 100 random partitions of the original data to generate a null distribution of tree scores with which to test the null hypothesis that the two data sets are random samples from a single statistical population. Maximum parsimony searches of the permutated data sets employed nearest-neighbor interchange branch swapping with the maximum number of saved trees set at 5000. Only potentially informative characters were included in the comparisons, and all were weighted equally.
Maximum parsimony analysis was implemented using full heuristic searches under the Fitch (unordered) criterion (Fitch, 1971). Starting trees for branch swapping were acquired by random addition, swapping on the best trees only when multiple starting trees existed. Stepwise addition sequence was simple, with one tree held at each step. The branch-swapping algorithm used was tree bisection-reconnection (TBR), swapping on best trees only with multiple trees saved. Multiple most parsimonious (MP) trees were combined as strict consensus trees. The rpl16 data set alone would not swap to completion, so that search was limited to 50 000 trees.
In addition to standard measures of fit of characters to the resultant trees such as consistency index (CI), retention index (RI), and rescaled consistency index, the strength of support for individual branches was estimated using jackknife values (Farris et al., 1996). Jackknife values are from 10 000 “full heuristic” searches with simple-sequence-addition replicates, TBR branch swapping with 36.8% character deletion, and emulating “Jac” resampling. Jackknife runs were performed using only informative characters in the data matrix, and jackknife consensus trees were constructed retaining branches with frequency of >50%. Additional tree topologies were examined for length using either topological constraint commands in PAUP* with 100 random addition sequences or with the tree editor in MacClade 4.0 (Maddison and Maddison, 2000). The significance of differences between constrained and unconstrained trees was tested using the Templeton (Wilcoxon signed-ranks; Templeton, 1983) nonparametric “Ttree Score” option in PAUP*.
RESULTS
Defining substitution and indel characters
In total, 4839 aligned nucleotide positions were considered in these analyses. No regions were excluded from the analyses because alignment was relatively straightforward. The number of nucleotide positions from each of the three chloroplast regions for a reference taxon (Brodiaea elegans) were as follows: 2035 from ndhF, 1004 from trnL-F, and 928 from rpl16. A total of 58 informative indels were coded: 3 from ndhF, 20 from trnL-F, and 35 from rpl16. Aligned nucleotides and indels gave a total of 4897 characters in the data matix for each taxon. Comparisons for alignment length and number of indels should not be drawn between the first two markers and rpl16, as rpl16 was not used in the Asparagales-wide survey. If it had been, it likely would have had a longer aligned nucleotide length and more indels. Interestingly, the supposedly “slowly evolving” ndhF region had in the 3′ half a comparable amount of potentially informative substitutions when compared with the “faster evolving” chloroplast intron and spacer regions, but ndhF had far fewer indels (Olmstead and Sweare, 1994; Kim and Jansen, 1995).
Phylogenetic reconstruction
Five separate analyses were carried out at two levels of taxonomic sampling. For the Asparagales-wide survey, ndhF and trnL-F are presented in a single combined analysis. For the Themidaceae survey, one analysis is presented for each of the three cpDNA regions, as well as a fourth analysis that combines the three cpDNA data sets. All analyses found a single island (sensu Maddison, 1991).
Asparagales-wide survey: ndhF and trnL-F
Because the results of Asparagales-wide surveys using the trnL-F region have been published recently (Meerow et al., 1999; Fay et al., 2000), only the combined ndhF and trnL-F analyses are presented here. The ILD test resulted in a value of 0.13, and because this was >0.05, we assumed that it was appropriate to combine data. Combining these ndhF and trnL-F resulted in 3581 characters, of which 576 were parsimony informative. Figure 2 is one of two MP trees (tree length = 1268, CI = 0.587, RI = 0.746). Note the high levels of jackknife support for five monophyletic clades: Themidaceae (100%), Hyacinthaceae (99%), Agavaceae + Anthericaceae (99%), Asparagaceae + Convallariaceae (99%), Agapanthaceae + Alliaceae + Amaryllidaceae (83%). Alliaceae and Amaryllidaceae have moderate support (85% jackknife) as sister groups. However, there is low support (<50% jackknife) for relationships between the clades.
Themidaceae survey: ndhF
The aligned ndhF matrix had 2295 total characters with 255 that were potentially parsimony informative (including three indels). Figure 3 is one of 48 MP trees (tree length = 524, CI = 0.613, RI = 0.869). Note the presence of four strongly supported monophyletic groups in this and the following analyses: (1) the Mexican genera that make up the Milla complex, with 99% jackknife support; (2) Brodiaea, Dichelostemma, and Triteleiopsis, with 97% jackknife support; (3) Triteleia, Bloomeria, and Muilla clevelandii, with 96% jackknife support; and (4) Androstephium and the other species of Muilla, with 100% jackknife support. The Milla complex has a sister relationship to the third and fourth clades, but it is weakly supported (50% jackknife support) and is not found in the remaining individual or combined analyses.
Themidaceae survey: trnL-F
The aligned trnL-F matrix had 1286 total characters with 169 that were potentially parsimony informative (including 20 indels). Figure 4 is one of eight MP trees tree length = 302, CI = 0.682, RI = 0.893). Four strongly supported monophyletic groups are found: (1) the Mexican genera that make up the Milla complex, with 97% jackknife support; (2) Brodiaea, Dichelostemma, and Triteleiopsis, with 97% jackknife support; (3) Triteleia and Bloomeria, with 95% jackknife support; and (4) Androstephium and three of the four species of Muilla, with 100% jackknife support. Muilla clevelandii is weakly supported (58% jackknife support) as being sister to the Muilla-Androstephium clade and not the Bloomeria-Triteleia clade, but this result is not found in the other individual or combined analyses.
Themidaceae survey: rpl16
The aligned rpl16 matrix had 1316 total characters with 212 that were potentially parsimony informative (including 35 indels). Figure 5 is one of the constrained 50 000 MP trees (tree length = 452, CI = 0.626, RI = 0.865). Four strongly supported monophyletic groups are found: (1) the Mexican genera that make up the Milla complex, with 99% jackknife support; (2) Brodiaea, Dichelostemma, and Triteleiopsis, with 98% jackknife support; (3) Triteleia, Bloomeria, and Muilla clevelandii, with 61% jackknife support; and (4) Androstephium and the other species of Muilla, with 100% jackknife support. The Milla complex has a highly supported sister relationship to the Triteleiopsis-Dichelostemma-Brodiaea clade, with 83% jackknife support and the presence of a 30 nucleotide long indel.
Themidaceae survey: cpDNA combined
Collectively, the three cpDNA data sets had 4897 total characters with 636 that were potentially parsimony informative (including 58 indels). Because the rpl16 data set would not run to completion, we could not conduct the ILD test using rpl16. However, we could conduct the ILD test for ndhF and trnL-F, which resulted in a value of 0.53. Given this result and the topological similarity between all three data sets, we assumed that it was appropriate to combine data. The same four clades found in the trnL-F analysis were found in all three analyses of single gene regions, and none of the clades that conflicted between two analyses were strongly supported. Figure 6 is one of eight MP trees (tree length = 1293, CI = 0.626, RI = 0.869). The same four strongly supported monophyletic groups are found: (1) the Mexican genera that make up the Milla complex, with 100% jackknife support; (2) Brodiaea, Dichelostemma, and Triteleiopsis, with 100% jackknife support; (3) Triteleia, Bloomeria, and Muilla clevelandii, with 96% jackknife support; and (4) Androstephium and the other species of Muilla, with 100% jackknife support. As for relationships among these four clades, the Androstephium-Muilla clade is well supported as sister to the Triteleia-Bloomeria-Muilla clevelandii clade, with 85% jackknife support. The Milla complex has a moderately supported sister relationship to the Brodiaea-Dichelostemma-Triteleiopsis clade, with 63% jackknife support. We tested the possibility of the Milla complex being sister to the other three main clades of the Themidaceae and found that this requires +3 extra steps in a topological contraints analysis (and was not significantly different from our consensus tree with the Templeton tests; P = 0.55). In the combined analysis, only four branches have jackknife support of <50%, so it is possible to characterize additional well-supported monophyletic groups within the four major clades. Although most of these relationships are also supported in the individual gene trees, only jackknife values for the combined analysis will be noted.
The Brodiaea-Dichelostemma-Triteleiopsis clade consists of six notable subclades (Fig. 6). Triteleiopsis palmerii is strongly supported as a member of one clade, with 100% jackknife support. Dichelostemma capitatum, a member of a second clade, is strongly supported (92% jackknife support) as sister to a third clade containing the other taxa of Dichelostemma and Brodiaea, with 100% jackknife support. The remaining four species of Dichelostemma form a highly supported fourth clade, with 96% jackknife support. Brodiaea orcuttii and B. filifolia are monophyletic in a fifth clade (100% jackknife support) and disjunct from a sixth clade consisting of the remainder of the Brodiaea species sampled (88% jackknife support).
Within the Milla complex, Milla appears to be polyphyletic, with 95% jackknife support for a clade containing Milla magnifica, Petronymphe, Jaimehintonia, and Dandya, 97% jackknife support for M. magnifica and Petronymphe, and 70% jackknife support for a clade containing Milla biflora, and a monophyletic Bessera (Fig. 6). However, while all of the genera of the Milla complex have been represented in these analyses, less than half of the species of Dandya and Milla have been sampled. Thus, any phylogenetic results pertaining to relationships within the Milla complex are tentative except for the non-monophyly of Milla.
Muilla clevelandii is strongly supported as belonging to a clade containing Bloomeria and Triteleia (Fig. 6; 96% jackknife support). With the current level of taxon sampling in Triteleia, the sister relationship of Triteleia and Bloomeria is well supported (100% jackknife support), as is the monophyly of Triteleia (92% jackknife support) and to a lesser degree the monophyly of Bloomeria (64% jackknife support). Within Triteleia, there are four clades all supported with 100% jackknife values: (1) T. montana and T. lemmonae; (2) T. peduncularis, T. laxa, and T. bridgesii; (3) T. hyacinthina and T. ixioides; and (4) T. grandiflora, T. crocea, and T. hendersonii.
Finally, with the removal of Muilla clevelandii, the other three species of Muilla form a monophyletic group (Fig. 6; 85% jackknife support) and Androstephium is supported as a monophyletic genus (92% jackknife support).
DISCUSSION
A major assumption in molecular systematics is that gene trees reflect species trees (Nei, 1987; Doyle, 1992; Avise, 1994; Maddison, 1995; Slowinski and Page, 1999). This assumption is particularly problematic when any single linkage group (e.g., plastid DNA) is the sole source of characters for phylogeny reconstruction because in the absence of other data it is impossible to determine whether even a robust plastid phylogeny is indicative of organismal lineages or reflects reticulating historical processes (e.g., hybridization, coalescence; reviewed in: Wendel and Doyle, 1998; Sytsma and Hahn, 2001). Furthermore, data incongruence is not necessarily a simple function of linkage (Naylor and Brown, 1998). However, while multiple data sets gathered from a single plastid can be strongly incongruent according to the partition homogeneity test, the individual gene regions can produce similar if not identical trees (Reeves et al., 2001). Fortunately, both levels of cpDNA studies conducted here have results consistent with those found in nuclear and in morphological data sets (Pires, 2000; Pires et al., 2001), suggesting that these particular plastid gene trees reflect to a large degree organismal lineages.
Phylogenetic assumptions made by biosystematists before 1993
Given the high degree of congruence between our cpDNA analyses (Figs. 2–6), we can now revisit the four assumptions made by biosystematists before 1993 regarding the phylogenetic relationships of Brodiaea and related petaloid monocots (Fig. 1). To review: (1) Brodiaea and relatives in western North America are related to taxa in the Alliaceae because they share an umbelliferous scape and a superior ovary; (2) Brodiaea and relatives are a monophyletic group separate from Alliaceae because they lack alliaceous chemistry and possess corms instead of tunicated bulbs for rootstock (see Table 1 for other distinctions); (3) Moore's two complexes, the Brodiaea complex (centered in the western United States) and the Milla complex (centered in Mexico), are both monophyletic groups; and (4) the Brodiaea complex consists of a monophyletic Brodiaea s.l. (= Brodiaea, Dichelostemma, and Triteleia) and allied genera.
First, although Brodiaea and relatives have been placed in several families, including Liliaceae (Cronquist, 1981), Amaryllidaceae (Traub, 1963; Niehaus, 1971; Keator, 1989), and Alliaceae (Dahlgren, Clifford, and Yeo, 1985), all the genera of Traub's subtribe Brodiaeinae have been presumed to be related to alliaceous taxa (Moore, 1953). Traub (1963, 1982) considered Brodiaea and relatives to be related to Alliaceae on the basis of sharing a superior ovary and scapose umbellate inflorescence and are thus separate from Liliaceae (superior ovary but no umbels) and Amaryllidaceae (umbels but inferior ovary).
As for the second assumption, Traub initially separated Brodiaea and relatives from other members of Alliaceae as subtribe Brodiaeinae (Traub, 1963), but later he elevated the group to family Brodiaeaceae (Traub, 1982). Brodiaeaceae was not published validly (Fay and Chase, 1996), but Traub's attempt to separate it further from Alliaceae illustrates that Brodiaea and relatives were consistently considered separate from, but closely related to, South American taxa of the Alliaceae.
The third assumption derived from Moore's (1953) infrafamilial classification that divided the 11 genera of tribe Brodiaeeae into two complexes on the basis of morphology and geography—the Milla complex and the Brodiaea complex (Fig. 1). The Milla complex (Bessera, Dandya, Milla, and Petronymphe) is centered in Mexico and defined as having corms with membranous tunics of minute parallel fibers and an ovary born on a gynophore that is generally adnate to the perianth tube. The Brodiaea complex (Androstephium, Bloomeria, Brodiaea, Dichelostemma, Muilla, Triteleia, and Triteleiopsis) is centered in western North America and defined as having corms with fibrous-reticulate tunics and an ovary that is either sessile or on a free gynophore. The monotypic Jaimehintonia B. Turner has been considered to be related to Androstephium (Turner, 1993), although other workers consider it closer to Dandya (T. M. H. Howard, personal communication). Moore's 1953 system was adopted by Traub (1963), and later by Niehaus (1971, 1980) and others (Keator, 1967, 1989). Thus, Moore's division of tribe Brodiaeeae into two complexes has remained largely unquestioned, even though the circumscription of some genera (e.g., Brodiaea) has varied markedly among authors.
The final persistent assumption, as mentioned previously, is that within Moore's 1953 Brodiaea complex, Brodiaea s.l. (Brodiaea, Dichelostemma, and Triteleia) was considered a separate taxonomic entity from the satellite genera Androstephium, Bloomeria, and Muilla (Fig. 1). The rationale for separating Brodiaea s.l. from the remaining genera has been based on tenuous morphological and biogeographic distinctions. Morphologically, species of Brodiaea s.l. have been said to share an extended perianth tube, while the other genera have been said to have either free tepals or a very short perianth tube (Keator, 1989). Biogeographically, species of Brodiaea s.l. are found primarily in northern California and the Pacific Northwest, whereas the rest of the Brodiaea complex is typically found in southern California and/or the Great Basin Floristic Province (Hoover, 1939a; Niehaus, 1971; Keator, 1989). Triteleiopsis is intermediate because it has a perianth tube but occurs in the Great Basin and in Baja California, and Hoover (1941) compared Triteleiopsis with both Brodiaea and Triteleia to identify it as a distinct genus. In addition, Hoover later considered Bloomeria another possible relative of Triteleia (Hoover, 1955). However, Niehaus (1980) and Keator (1989) elaborated only on Hoover's (1939a, b) three-genera concept and ignored Hoover's later observations that Triteleiopsis may be affiliated with either Brodiaea or Triteleia and that Bloomeria may be related to Triteleia.
Suprafamilial relationships of Themidaceae
Although the first two assumptions regarding whether Themidaceae is monophyletic and related to Alliaceae have been previously addressed (Fay and Chase, 1996; Meerow et al., 1999; Fay et al., 2000; Pires et al., 2001), we wanted to confirm those results by (1) using a wider taxonomic sampling that included all the genera of the Themidaceae and (2) applying a molecular marker, ndhF, that had not been previously used to investigate relationships within the higher Asparagales. As Fig. 2 demonstrates, Alliaceae is closely related to Amaryllidaceae and Agapanthaceae, so Alliaceae does not have a direct relationship to Themidaceae. We also found Themidaceae to be monophyletic, a significant result because it allows us to evaluate lineages within the Themidaceae with greater certainty. In this study, Hyacinthaceae is monophyletic and sister to Themidaceae. However, the assumption that Themidaceae is sister to Hyacinthaceae has been recently challenged by Fay et al. (2000), who surveyed all but one of the families (Hesperocallidaceae) currently recognized in the Asparagales sensu Angiosperm Phylogeny Group (1998) with a four-plastid-region data set. Fay et al. (2000) found Themidaceae to be strongly monophyletic, but the sister relationship to Hyacinthaceae was destabilized with additional taxa (most notably the taxonomically isolated Mediterranean genus Aphyllanthes). This result raises the possibility that Themidaceae is sister to a different clade that could include representatives from several other families in the higher Asparagales, including Agavaceae, Anthericaceae, Asparagaceae, Behniaceae, Convallariaceae, Herreriaceae, and Laxmanniaceae. However, McPherson and Graham (2001) found Themidaceae to be sister to Hyacinthaceae and Aphyllanthaceae in a phylogenetic study using 17 chloroplast genes. This finding is consistent with other recently published results of studies that sampled more widely across the petaloid monocots using both chloroplast markers (Meerow et al., 1999; Pfosser and Speta, 1999; Chase et al., 2000) and nuclear markers (Chase et al., 2000; Pires, 2000 and unpublished data) and was the basis for selecting Hyacinthaceae as the primary outgroup for our analyses. Fay and Chase (1996) indicated that the presence of a corm and multiple bracts subtending the inflorescence are synapomorphies for the Themidaceae. Furthermore, they stated that the relationship of Themidaceae to Hyacinthaceae is supported by the presence in both groups of bracteate pedicels and anatropous ovules, as well as the plesiomorphic characters of crassinucellate ovules and hollow styles.
Several recent studies have found strong support for Alliaceae as being sister to Agapanthaceae and Amaryllidaceae (Meerow et al., 1999; Fay et al., 2000; McPherson and Graham, 2001). If the clade containing those three families continues to be phylogenetically distant to the Themidaceae, we could conclude that the umbelliferous inflorescence used as a morphological character to unite Brodiaea and relatives with taxa in the Alliaceae is problematic because the presence of an umbel may have evolved independently several times in the Asparagales (Themidaceae, Asparagaceae, and the Alliaceae-Amaryllidaceae-Agapanthaceae clade). Furthermore, the separation of Themidaceae from South American taxa of Alliaceae (e.g., the South American “brodiaeas”; Table 1) and the potential relationship of Themidaceae to alternative clades (e.g., an Old World Hyacinthaceae radiating out of South Africa) suggest novel biogeographic origins for Brodiaea and relatives. Despite the lack of resolution among families in the higher Asparagales, all these recent analyses continue to strongly support a monophyletic Themidaceae, allowing the possibility of exploring the relationships within the family using a variety of higher Asparagales as a global outgroup (Maddison, Donoghue, and Maddison, 1984).
Phylogenetic relationships within Themidaceae
Assessing the validity of the first two assumptions was primarily an exercise in verification, but testing the third and fourth assumptions (see Discussion: Phylogenetic assumptions made by biosystematists before 1993) brings us into new territory (Pires et al., 2001) and is the primary focus of this paper. Given the congruence of the individual gene trees (Figs. 3–5) with each other and the combined analysis (Fig. 6), we will use the combined analysis as the basis for the following discussion. Although Fig. 6 is derived exclusively from plastid sources, the results from nuclear (internal transcribed spacer region) and morphological analyses (Pires, 2000; Pires et al., 2001) are consistent with the results presented here and indicate that the four major clades under discussion are robust, not just when analyzed with a wide variety of phylogenetic methods but also when nonplastid data are used. Thus, we assume that Fig. 6 represents not only the plastid lineage but also our best current representation of organismal histories. We will use the topology of Fig. 6 to evaluate the evolution of morphological characters and biogeographic distribution patterns (Fig. 7). In light of the molecular phylogenies, we will address four clades: (1) the monophyletic Milla complex; (2) Brodiaea, Dichelostemma, and Triteleiopsis; (3) Triteleia, Bloomeria, and Muilla clevelandii; and (4) Androstephium and the other species of Muilla. These findings are significant because the last three clades are in conflict with previous assumptions (Fig. 1).
The Milla complex clade
The genera in Moore's (1953) Milla complex (Bessera, Dandya, Milla, and Petronymphe) were described as members of Hutchinson's (1934) Allieae with membranous tunicated corms and gynophores that are adnate to the floral tube along the entire circumference or along three ridges opposite the outer tepals. Our results (Figs. 3–6) support the monophyly of Moore's complex, with two notable additions to the clade: (1) Jaimehintonia, which Turner had associated with Androstephium, and (2) Petronymphe, which had previously been excluded in error (Fay and Chase, 1996). Morphological synapomorphies for the clade include not only the corm characters and the fusion of the basalmost portion of the ovary to the floral tube, as described by Moore (1953), but also the presence of a stamen filament tube. While the corm characters appear to be unique to the Milla complex within the Themidaceae, the gynoecial and androecial characters are found in nearby clades to some degree. The length of the ovary stipe within the Milla complex varies, as does the degree of fusion between the stipe and the perianth tube. While Dandya has a short ovary stipe that is only basally adnate to the perianth tube, Milla and Bessera have spectacularly elongated ovary stipes that are adnate along three longitudinal lines for almost the entire length of the perianth tubes. Androstephium is sometimes reported as having a portion of its ovary adnate to the perianth, but it lacks an ovary stipe. Conversely, Triteleia possesses an ovary stipe, but it is never fused to the perianth tube (Fig. 7).
The androecial patterns differ between the Milla and Brodiaea complexes. Stamen filaments of most members of the Milla complex (with the exception of Milla itself) are attached at the base of the perianth, whereas evolutionary diversification in the Brodiaea complex has resulted in the attachment of stamen filaments to the mid- to upper portion of the perianth in most members. In addition, some members of the Milla complex have a tube formed from the connation of the free portions of stamen filaments. These results are consistent with the biogeographic distinctiveness of the genera of the Milla complex, which are found exclusively in Mexico, except for M. biflora, which ranges into southern Arizona, southern New Mexico, and Guatemala (Fig 7).
The Brodiaea, Dichelostemma, and Triteleiopsis clade
Our results strongly support a novel Brodiaea-Dichelostemma clade comprising Brodiaea and most of Dichelostemma that is sister to D. capitatum and Triteleiopsis palmerii (Figs. 3–6). The Triteleiopsis-Dichelostemma-Brodiaea clade is notable because (1) Brodiaea and Dichelostemma are not sister to Triteleia, as assumed by the biosystematic framework (Fig. 1, Table 2) that has guided traditional systems of classification; (2) Dichelostemma is not monophyletic unless one excludes D. capitatum; and (3) Triteleiopsis palmerii is not closely related to Triteleia.
Historically, Brodiaea has been kept separate from Dichelostemma on the basis of chromosome number, hybridization studies, and a few morphological characters. Brodiaea has a straight scape, open umbels, terete leaves, and a distinctly three-lobed stigma, Dichelostemma has a lax scape, dense umbels, flattened keeled leaves, and a weakly lobed stigma. However, Brodiaea and Dichelostemma are similar morphologically (and differ from Triteleia) in that they share three basifixed anthers, a three-lobed stigma, basal ovary attachment, and hypogeal germination (vs. six versatile anthers, unlobed stigma, stipitate ovary attachment, and epigeal germination in Triteleia; see Fig. 7). Given this, it is perhaps not surprising that the Brodiaea-Dichelostemma clade is found in all of the analyses (Figs. 2–6) and is consistent with Hoover's (1940) suggestion that the two genera were more closely related to each other than to any other group. Hoover supported his hypothesis of a close relationship between Brodiaea and Dichelostemma with androecial characters, and in Pires et al.'s (2001) study the relationship of these two genera is supported primarily by their synapomorphic possession of three fertile stamens opposite the inner tepals and sterile appendages opposite the outer tepals. Finally, the three stamens of Brodiaea and Dichelostemma are closely appressed to the style at anthesis, whereas in the other taxa the six fertile stamens stand apart from the style.
Dichelostemma capitatum has long been a “problematic species” (Keator, 1992, p. 24) because unlike all other species of Brodiaea and Dichelostemma, which have hypogeal germination and three fertile stamens opposite the inner tepals D. capitatum has epigeal germination and six fertile stamens (three tall stamens opposite the inner tepals and three short stamens opposite the outer tepals; Keator, 1967, 1992; Lenz, 1976). However, embryological evidence (Berg, 1996) and data from other molecular markers (Pires et al., 2001) show that this species is sister to the Brodiaea-Dichelostemma clade. Dichelostemma capitatum is also a wide-ranging species (across western North America), whereas the other members of Brodiaea and Dichelostemma have more isolated distributions, primarily in California and Oregon (Fig. 7).
The monotypic Triteleiopsis was originally segregated from Triteleia on the basis of numerous vegetative and floral characters (Hoover, 1941). Hoover described leaves of Triteleiopsis as originating from the stem above the corm, whereas the leaves of Triteleia originate directly from the corm. Triteleiopsis is found in the dune habitats of the Sonoran Desert, primarily in Mexico (northern Baja California and Sonora) but creeping into the southwest corner of Arizona (Fig. 7).
The Brodiaea-Dichelostemma-Triteleiopsis clade appears to show an evolutionary trend in androecial evolution (Fig. 7). Members of the entire clade have basifixed anthers. Basifixed anthers are also found in Androstephium, but all other members of the Themidaceae have versatile anthers. Within the Brodiaea-Dichelostemma-Triteleiopsis clade there is a reduction in androecial features: (1) Triteleiopsis has six fertile equally tall stamens; (2) D. capitatum has three tall fertile stamens opposite the inner perianth whorl and three reduced fertile stamens opposite the outer perianth whorl; and (3) the remaining four species of Dichelostemma and Brodiaea have only three fertile stamens opposite the inner perianth whorl and sterile staminodes in the outer whorl (staminodia sensu Hoover, 1939b, 1940; staminodia or elaborations of filament appendages sensu Keator, 1967; staminodia, sterile tissue derived from adjacent tepals, or staminal tube elaborations sensu Lenz, 1976). The sterile staminodes are reduced or lost altogether in two monophyletic groups: a clade containing Brodiaea orcuttii and B. filifolia and a clade containing Dichelostemma ida-maia, D. multiflorum, and D. congestum. In addition to these morphological changes, there is a disjunction in biogeographic distribution (Fig. 7) within the Brodiaea-Dichelostemma-Triteleiopsis clade. While Triteleiopsis and the wide-ranging D. capitatum have overlapping distributions in Arizona and Baja California, the remaining species occur primarily in California, with B. coronaria and D. congestum extending into northern Oregon and western Washington.
The Triteleia-Bloomeria-Muilla clevelandii clade
Our results strongly support a novel clade consisting of Triteleia, Bloomeria, and Muilla clevelandii (Figs. 3–6). This is significant for two reasons. First, Triteleia has traditionally been considered related to Brodiaea and Dichelostemma and separate from Bloomeria and other genera (Fig. 1, Table 2) on the basis of a single shared character: an extended perianth tube (Keator, 1989). Our results strongly indicate that this is not the case and that Triteleia is closely related to Bloomeria, Muilla, and Androstephium. That Bloomeria and Triteleia are closely related is not surprising in the context of historical debates (Hoover, 1941, 1955). The close relationship of Bloomeria and Triteleia was suggested by Hoover (1955), who emphasized that the two genera had similarities in several morphological structures (corm, leaf, anthers, stamen appendages, scape pubescence, seeds) as well as the same base chromosome number. Furthermore, Hoover (1955) rejected the perianth tube distinction between Brodiaea, Dichelostemma, and Triteleia on the one hand and Androstephium, Bloomeria, and Muilla on the other, by pointing out that the perianth segments of some Bloomeria are slightly joined at the base and that T. ixioides has a very short perianth tube. In retrospect, it is unfortunate that Niehaus (1971, 1980) and Keator (1967, 1989) followed Hoover's original (1939a, b) definition of Brodiaea s.l. (Brodiaea, Dichelostemma, and Triteleia) and ignored his later observations about the possible relationships of Bloomeria and Triteleiopsis to these three core genera (Hoover, 1941, 1955). However, Hoover himself maintained Bloomeria as a separate genus from Triteleia because Triteleia has an ovary stipe that Bloomeria lacks and because Triteleia is rarely found south of the Tehachapi and Santa Lucia Mountains in California, where Bloomeria occurs (Hoover, 1955). Triteleia is widely distributed west of the Rocky Mountains, but its greatest diversity is in the Klamath area of northwestern California and southwestern Oregon.
The second reason why the Triteleia-Bloomeria-Muilla clevelandii clade is significant, of course, is the novel placement of M. clevelandii. Muilla clevelandii is another problematic species because it was transferred from Bloomeria to Muilla by Hoover (1955). Ingraham's (1953) differentiation of Bloomeria (including Bloomeria clevelandii) from Muilla emphasized the presence of jointed pedicels in the former and the lack of jointed pedicels in the latter. In his “Key to Bloomeria, Muilla and Closely Related Genera,” these pedicel characters unite Bloomeria with Brodiaea and Muilla with Allium. However, Hoover (1955, p. 23) argued that the presence or absence of jointed pedicels to define Bloomeria from Muilla was “difficult to apply to actual specimens” and that instead, “Muilla differs consistently from Bloomeria in the larger number of leaves which are neither keeled nor channeled (in this respect resembling true Brodiaea rather than either Bloomeria or Triteleia).” Hoover's only reservation in naming M. clevelandii was that it shared yellow-colored flowers with Bloomeria, whereas the flowers of the other Muilla species are white.
Considering the overall trends in the Triteleia-Bloomeria-Muilla clevelandii clade, there are no obvious morphological synapomorphies that unite the group as a whole, although embryological characters are particularly promising (Berg, 1996). Triteleia is well defined by the presence of an ovary stipe that is not found in any of the genera of the Themidaceae outside of the Milla complex and a few species of Dichelostemma (Fig. 7). Biogeographically, Bloomeria and Muilla clevelandii are found along the southern California coast, while Triteleia has its center of diversity in northern California and the Pacific Northwest, with T. lemmonae found only on the Mogollon Plateau of Arizona and T. grandiflora and T. hyacinthina ranging east into Montana and Wyoming (Fig. 7).
The Androstephium-Muilla clade
This clade is significant because while Bloomeria and Triteleiopsis have at least been suggested to have affinities with Triteleia and Brodiaea, few phylogenetic relationships have previously been hypothesized for either of these genera (Fig. 1). The fact that Muilla is an anagram of Allium is suggestive of phylogenetic assumptions, as is Ingraham's (1953) key that united Muilla to Allium and Bloomeria to Brodiaea based on pedicel characters. Androstephium has been considered intermediate between the Brodiaea and Milla complexes because it has a corm tunic that is perhaps not as fibrous-reticulate as that in most members of the Brodiaea complex. Interestingly, Moore's (1953) placement of Androstephium in his Brodiaea complex contradicted Watson's (1879) alliance of it with Bessera. Although not noted by Moore (1953), Androstephium also has the basalmost portion of the ovary fused to the floral tube, a characteristic of the Milla complex. Turner (1993) suggested that his newly described Jaimehintonia may have affinities with Androstephium. Morphologically, neither of these genera are well known with the exception of a tantalizingly brief embryological comparison of Muilla to Allium (Berg and Maze, 1966). Recent generic and species recircumscriptions (Shevock, 1984), recent range extensions (White, Sanders, and Wilcox, 1996), and few chromosome counts (Lenz, 1975) reflect the fact that these two satellite genera have historically been ignored by biosystematic workers, and as a result few morphological or biogeographic (Fig. 7) conclusions can be drawn.
Conclusion
The taxonomic history of Brodiaea and relatives has entailed four implicit phylogenetic assumptions: (1) the Alliaceae are separate from other petaloid plant families on the basis of both having an umbel and a superior ovary; (2) a North American group of taxa are separate from a South American group of alliaceous taxa (Table 1); (3) the North American taxa are split into a Mexican-centered Milla complex, which is separate from a western United States-centered Brodiaea complex; and (4) Brodiaea s.l. (Brodiaea, Dichelostemma, and Triteleia) are separate from the other genera in the Brodiaea complex (Fig. 1, Table 2). Recent studies and the results reported here indicate that the first two assumptions are only partially correct. Although Brodiaea and relatives form a strongly supported monophyletic group, Themidaceae does not appear to be directly related to Alliaceae but rather to Hyacinthaceae and other families in the higher asparagoids of the Asparagales (Fay and Chase, 1996; Meerow et al., 1999; Fay et al., 2000; Pires et al., 2001). The combination of possessing an umbelliferous scape and a superior ovary probably evolved more than once in the petaloid monocots, a possibility that opens many new prospects for the biogeographic origins of the New World Themidaceae.
The primary significance of this study is that four novel clades are described that deeply challenge previous phylogenetic assumptions. In this study, the Milla complex is monophyletic if recircumscribed to include Jaimehintonia and Petronymphe, in contrast with previous studies (Turner, 1993; Fay and Chase, 1996). Surprisingly, these results consistently show the paraphyly of the Brodiaea complex in the MP trees (Figs. 3–6). As noted in RESULTS, there is only very weak support for the areas of disagreement between the individual gene trees (e.g., Milla complex is sister to the Androstephium-Muilla-Bloomeria-Triteleia clade in the ndhF analysis, and Muilla clevelandii is sister to the rest of Muilla and Androstephium in the trnL-F analysis). A paraphyletic Brodiaea complex containing a monophyletic Milla complex suggests that the radiation into Mexico and the morphological characters associated with the Milla complex occurred later in the evolution of the Themidaceae than has been previously suggested (Fig. 7).
We found that the fourth assumption concerning the monophyly of Brodiaea s.l. (Brodiaea, Dichelostemma, and Triteleia) and the presumed monophyly of the group containing the remaining satellite genera of the Brodiaea complex (Androstephium, Bloomeria, Muilla, and Triteleiopsis) is deeply misleading (Fig. 6). This finding is significant because many biosystematic comparisons that occurred in the latter half of the 20th century assumed these relationships (Fig. 1, Table 2). Similarly, taxonomic disagreements concerning the proper generic limits of Brodiaea, Dichelostemma, and Triteleia were limited to two schools of thought: either a single genus (Brodiaea s.l.) with three subgenera (Watson, 1879; Abrams, 1923; Jepson, 1925; Munz, 1959) or three separate genera (Greene, 1886; Hoover, 1939a; Keator, 1967, 1989, 1993a–c; Niehaus, 1971, 1980). We have shown that both these alternatives are incorrect because Brodiaea s.l. is polyphyletic when the satellite genera are considered. Thus, the observation that Brodiaea, Dichelostemma, and Triteleia share an extended perianth tube and a similar biogeographic distribution is problematic precisely because the extended perianth tube presumably has evolved at least twice in the Themidaceae and because there appears to have been two radiations into the California-Pacific Northwest area (Fig. 7). Future studies (Pires, 2000 and unpublished data) will use nuclear-based DNA markers (internal transcribed spacer region) and morphological data to verify this phylogenetic framework inferred from three cpDNA data sets. More important, these results invite systematists to revisit the rich biosystematic research efforts concerning the floral evolution, polyploidy, and patterns of endemism found in Brodiaea, Dichelostemma, and Triteleia and to extend those studies to include the other genera in the Themidaceae.
Assumed phylogenetic relationships of Brodiaea and related petaloid monocots before 1993.
One of two trees found in the maximum parsimony analysis using ndhF and trnL-F in Asparagales-wide survey (tree length = 1268, consistency index [CI] = 0.587, retention index [RI] = 0.746). Numbers above branches indicate branch length (ACCTRAN optimization), and numbers below branches indicate jackknife values >50%. Branches drawn with dashed lines were not found in the strict consensus tree. Arrows indicate clades discussed in the text.
One of 48 trees found in the maximum parsimony analysis of Themidaceae using ndhF (tree length = 524, CI = 0.613, RI = 0.869). See Fig. 2 for meaning of numbers, dashed lines, and arrows
One of eight trees found in the maximum parsimony analysis of Themidaceae using trnL-F (tree length = 302, CI = 0.682, RI = 0.893). See Fig. 2 for meaning of numbers, dashed lines, and arrows
One of 50 000 trees (constrained) found in the maximum parsimony analysis of Themidaceae using rpl16 (tree length = 452, CI = 0.626, RI = 0.865). See Fig. 2 for meaning of numbers, dashed lines, and arrows.
One of eight trees found in the maximum parsimony analysis of Themidaceae using all three cpDNA regions (tree length = 1293, CI = 0.626, RI = 0.869). See Fig. 2 for meaning of numbers, dashed lines, and arrows. Note that the Brodiaea s.l. (Brodiaea, Dichelostemma, and Triteleia) is not monophyletic and that the Milla complex is embedded in the Brodiaea complex
Morphological characters and biogeography mapped onto the cpDNA-based tree of Themidaceae of Fig. 6. Note that the extended perianth tube used to define Brodiaea s.l. has evolved independently twice and that there have been multiple radiations into California and the Pacific Northwest
