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

This Article Abstract Full Text (PDF) Submit a response Alert me when this article is cited Alert me when eLetters are posted Alert me if a correction is posted Services Email this article to a friend Similar articles in this journal Similar articles in ISI Web of Science Alert me to new issues of the journal Download to citation manager Citing Articles Citing Articles via ISI Web of Science (1) Citing Articles via Google Scholar Google Scholar Articles by Douglas, N. A. Articles by Manos, P. S. Search for Related Content PubMed Articles by Douglas, N. A. Articles by Manos, P. S. Agricola Articles by Douglas, N. A. Articles by Manos, P. S.

Molecular phylogeny of Nyctaginaceae: taxonomy, biogeography, and characters associated with a radiation of xerophytic genera in North America¹

Department of Biology, Duke University, Box 90338, Durham, North Carolina 27708 USA

Received for publication June 22, 2006. Accepted for publication February 28, 2007.

ABSTRACT

The four o'clock family (Nyctaginaceae) has a number of genera with unusual morphological and ecological characters, several of which appear to have a "tendency" to evolve repeatedly in Nyctaginaceae. Despite this, the Nyctaginaceae have attracted little attention from botanists. To produce a phylogeny for the Nyctaginaceae, we sampled 51 species representing 25 genera (of 28–31) for three chloroplast loci (ndhF, rps16, rpl16, and nrITS) and included all genera from North America. Parsimony, likelihood, and Bayesian methods were used to reconstruct the phylogeny for the family. The family is neotropical in origin. A radiation of woody taxa unites Pisonia and Pisoniella with the difficult tropical genera Neea and Guapira, which also form a clade, though neither appears to be monophyletic. This group is sister to a clade containing Bougainvillea, Belemia, and Phaeoptilum. A dramatic radiation of genera occurred in the deserts of North America. The tribe Nyctagineae and its subtribes are paraphyletic, due to over-reliance on a few homoplasious characters, i.e., pollen morphology and involucre presence. Two notable characters associated with the desert radiation are cleistogamy and edaphic endemism on gypsum soils. We discuss evolutionary trends in these traits in light of available data about self-incompatibility and gypsum tolerance in Nyctaginaceae.

Key Words: biogeography • cleistogamy • gypsophily • homoplasy • mating system • Nyctaginaceae • phylogeny • pollen morphology

Nyctaginaceae Juss. is a family of 28–31 genera and 300–400 species, that contains the familiar cultivated four o'clocks (Mirabilis jalapa) and bougainvillea (Bougainvillea spp.). Nyctaginaceae has long been known to be one of the core groups of families of Caryophyllales (Centrospermae) on the basis of the presence of betalain pigments, free-central placentation, p-type sieve tube elements, and the presence of perisperm, as well as molecular evidence (Bittrich and Kühn, 1993 ; Bremer et al., 2003 ). Within this group, the modern consensus is that Nyctaginaceae are closely related to certain monocarpellate members of a paraphyletic Phytolaccaceae, especially subfam. Rivinoideae (Rodman et al., 1984 ; Rettig et al., 1992 ; Downie and Palmer, 1994 ; Behnke, 1997 ; Downie et al., 1997 ; Cuenoud et al., 2002 ), although Sarcobatus (Sarcobataceae) has also been implicated as a close relative of this group (Behnke, 1997 ; Cuenoud et al., 2002 ).

Nyctaginaceae have a uniseriate petaloid perianth, usually interpreted as sepalous in origin (Rohweder and Huber, 1974 ). In most taxa the lower part of the perianth is fleshy or coriaceous and encloses the superior ovary, giving it the appearance of an inferior ovary. This accessory fruit is persistent and accrescent around the mature achene. While technically a diclesium (Bogle, 1974 ; Spellenberg, 2003 ), it is typically referred to as an "anthocarp."

Most genera can be recognized on the basis of fruit structure alone. In Boldoa, Cryptocarpus, and Salpianthus, the perianth is persistent but not accrescent, and thus these taxa lack the anthocarp (Bittrich and Kühn, 1993 ). In Andradea, Leucaster, and Reichenbachia, the perianth is variously accrescent but is not expanded (Bittrich and Kühn, 1993 ). However, in the remaining genera the anthocarp completely encloses the fruit and takes many forms (Willson and Spellenberg, 1977 ; Bittrich and Kühn, 1993 ). In taxa in which anthocarps are ribbed, the 3–10 ribs can be elaborated into wings (Phaeoptilum, Grajalesia, Tripterocalyx, Abronia, and some Colignonia, Acleisanthes, and Boerhavia), covered by viscid glandular hairs or warts (Pisonia, Pisoniella, Cyphomeris, Commicarpus, and some Boerhavia and Acleisanthes), or unelaborated, to leave an essentially gravity-dispersed fruit (Mirabilis, Anulocaulis, Nyctaginia, and some Colignonia and Boerhavia). Fleshy anthocarps are probably bird-dispersed in Neea and Guapira. They are also found in Okenia, though this genus is geocarpic and the seeds generally germinate at the spot where they are "planted" by the maternal individual (N. Douglas, personal observation). The unusual anthocarps of Allionia are boat-shaped, with two rows of inward-pointing teeth lining the concave side, suggesting possible exozoochory or wind dispersal, though no observations on this are available. In herbaceous taxa, at least, species-level characters are often found in this structure (Willson and Spellenberg, 1977 ; Spellenberg, 2003 ).

The family was treated by Heimerl in Die Natürlichen Pflanzenfamilien (Heimerl, 1889 , 1934 ) and by Standley in several papers (Standley, 1909 , 1911 , 1918 , 1931a , b ) by which time most of the currently recognized genera had been described. Standley (1931a) formally transferred Oxybaphus L'Hér. ex Willd., Hesperonia Standl., Quamoclidion Choisy, and Allionella Rydb. into Mirabilis, though this has been overlooked in some floras (e.g., Kearney and Peebles, 1960 ). Heimerl (1934) synthesized the family as it was known, including in his classification genera that had been recently described by Standley (i.e., Pisoniella, Cuscatlania). He based his supergeneric classification on a combination of plant habit, indumentum, linear vs. capitate stigma, straight vs. curved embryo, sex distribution, pollen grain morphology, and the occurrence of bracts or involucre (Bittrich and Kühn, 1993 ; Heimerl, 1934 ). Bittrich and Kühn (1993) provided the most recent summary of the classification at the tribal and subtribal level (Table 1). Their treatment broadly followed that of Heimerl (1934) , adjusting ranks and incorporating genera described after 1934, i.e., Caribea. It recognized six tribes, two of which, Pisonieae and Nyctagineae, contain the majority of genera and species (Table 1, Pisonieae: six genera, ca. 200 spp.; Nyctagineae: 14 genera, ca. 100 spp.).

Two major centers of distribution have been noted for the Nyctaginaceae (Standley, 1909 ). The first is in the neotropics and Caribbean, characterized by arborescent genera such as Neea, Guapira, Pisonia, and Bougainvillea, as well as the herbaceous Colignonia and Salpianthus. The second is in arid western North America, where several herbaceous or suffrutescent genera are native, including Boerhavia, Mirabilis, Abronia, Acleisanthes sensu Levin (2002) , and Commicarpus. A few genera are widespread in tropical and subtropical regions of the world (Boerhavia, Commicarpus, Pisonia): Mirabilis is present in North and South America with one species in Asia, and Acleisanthes contains the disjunct A. somalensis from Somalia. Mirabilis (M. jalapa, M. oxybaphoides) and Bougainvillea (B. glabra, B. spectabilis, B. peruviana, and numerous hybrid cultivars) are naturalized in many parts of the world. Only one genus is restricted to the Old World, the monospecific Phaeoptilum of southwestern Africa.

The first molecular phylogenetic study of Nyctaginaceae was presented by Levin (2000) . The focus was on species in certain genera of tribe Nyctagineae sensu Bittrich and Kühn (1993) , including genera in subtribes Nyctagininae (Allionia, Mirabilis) and Boerhaviinae (Acleisanthes, Selinocarpus, Boerhavia), as well as Abronia and Pisonia. The study justified the formal combination of Acleisanthes, Selinocarpus, and Ammocodon (Levin, 2002 ; Spellenberg and Poole, 2003 ), but due to limited sampling of genera, it was not possible to evaluate the monophyly of the subtribes of Nyctagineae (Levin, 2000 ). The Flora of North America treatment of Nyctaginaceae (Spellenberg, 2003 ), while not referring to tribal classification, reflected these and other taxonomic changes for the genera and species that occur in North America north of Mexico (Table 1).

In the herbaceous taxa of Nyctaginaceae found in the deserts of North America, several unusual characters occur with notable frequency. As indicated by the common name for the family, species in several genera (Anulocaulis, Cyphomeris, Acleisanthes, Mirabilis, Abronia, and Tripterocalyx) flower in the evening and are adapted to moth pollination (Baker, 1961 ; Grant, 1983 ; Grant and Grant, 1983 ; Hernández, 1990 ; Hodges, 1995 ; Levin et al., 2001 ). Internodal bands of viscid secretions, which may discourage aphid colonization (McClellan and Boecklen, 1993 ), are present in Anulocaulis, Cyphomeris, and some species of Boerhavia. As mentioned, anthocarp morphology is also variable, with wings and viscid glands being common modifications.

Because these characters are often polymorphic at the generic level, they would seem to represent evolutionary "tendencies." Sanderson (1991) discussed evolutionary tendencies in explicit phylogenetic terms: a tendency is a concentrated distribution of homoplasy within a tree. The main objection to the study of tendencies is the difficulty in defining the taxonomic scope at which they operate, in other words, it is "... biologically inappropriate [when investigating a hypothesized tendency] to include taxa that cannot under any circumstances exhibit the states of interest" (Sanderson, 1991 , p. 357). Thus, when considering whether a character has a tendency to evolve, it is first necessary to evaluate the range of taxa in which it could potentially appear. In some cases, it may be possible to identify another character upon which the evolution of the character of interest is dependent. If this other trait is itself uniquely derived, its occurrence will define the group in which the tendency may conceivably exhibit itself. If the independent character is itself derived multiple times, then the problem is pushed back so that the challenge is first to explain the tendency for the independent character to evolve in the group.

In the case of tendencies in Nyctaginaceae, it is not immediately obvious what sorts of traits may be required to enable, for instance, a shift to nocturnal pollination or the development of viscid bands on stem internodes. There are two traits, however, that seem to have a tendency to evolve in Nyctaginaceae and that we can reasonably assume are contingent on other traits: the evolution of cleistogamy is improbable without prior self-compatibility, and lineages that specialize on gypsum are unlikely to have arisen from lineages with no latent or expressed gypsum tolerance.

Cleistogamous (closed, self-fertilizing) flowers are produced in addition to chasmogamous (open) flowers in four genera of Nyctaginaceae: Acleisanthes, Cyphomeris, Nyctaginia, and some Mirabilis (Cruden, 1973 ; Spellenberg and Delson, 1974 ; Fowler and Turner, 1977 ; Levin, 2002 ). Though species with cleistogamous flowers have evolved in a number of angiosperm families, only in much larger families, e.g., Poaceae, Fabaceae, and Malpighiaceae, is this trait found in as many genera (Lord, 1981 ). Despite a long awareness of this phenomenon generally (Darwin, 1884 ), the evolution of this character has only rarely been investigated with phylogenetic methods (Desfeux et al., 1996 ; Bell and Donoghue, 2003 ).

Second, as in many caryophyllid families, e.g., Amaranthaceae and Portulacaeae, there is a propensity in many Nyctaginaceae to be tolerant of, or specialists of, gypseous soils. Outcrops of gypsum (hydrous calcium sulfate) are quite common in arid North America, especially in the Chihuahuan Desert. These areas have a flora characterized by gypsophiles, which never occur on other substrates, and gypsum-tolerant species, which are found on both gypseous and nongypseous soils (Waterfall, 1946 ; Parsons, 1976 ; Meyer, 1986 ). In the United States and Mexico, Nyctaginaceae are well represented in gypsum communities (Parsons, 1976 ). At least 25 species in seven genera are known to occur on gypsum. Of these, roughly half are known gypsophiles, found only on gypsum soils (Johnston, 1941 ; Waterfall, 1946 ; Fowler and Turner, 1977 ; Turner, 1991 , 1993 ; Spellenberg, 1993 , 2003 ; Mahrt and Spellenberg, 1995 ; Harriman, 1999 ; Levin, 2002 ).

Although gypsum soils support a distinct flora, the evolution of gypsophily is not understood as well as other cases of edaphic endemism. Gypsum is not an inherently poor substrate for plants in the same way as soil with, for instance, toxic levels of heavy metals (Cockerell and Garcia, 1898 ; Johnston, 1941 ; Loomis, 1944 ; Parsons, 1976 ; Meyer, 1986 ; Oyonarte et al., 2002 ). Recent experimental work has pointed toward mechanical, rather than chemical, factors to explain the limited flora of gypsum soils: seedlings of nongypsophiles are unable to penetrate the hard crust typical of gypseous soils. This indicates that adaptations of gypsum-tolerant taxa primarily act to enhance survival in the establishment stage (Meyer, 1986 ; Meyer et al., 1992 ; Escudero et al., 1997 , 1999 , 2000 ; Romao and Escudero, 2005 ).

Edaphic-endemic species are sometimes found to be related to species that are merely tolerant: in the case of a serpentine endemic species of Layia (Asteraceae), certain populations of a non-endemic progenitor species were found to tolerate serpentine soils (Baldwin, 2005 ). Thus, even in the case of highly toxic soils, saltational speciation (Antonovics, 1971 ; Kruckeberg, 1986 ) is not required to explain edaphic endemism. These lines of evidence, and the fact that roughly half of the species of Nyctaginaceae found on gypsum are not restricted to it, make it reasonable to assume that an underlying ability to survive in gypsum soils is an early stage in the evolution of this type of edaphic endemism in Nyctaginaceae.

In principle, for both of these examples, the evolution of both the independent and contingent characters can be reconstructed on a phylogeny. With an understanding of the distribution of homoplasy in Nyctaginaceae, we will have a more robust framework for asking questions about character evolution and adaptation to xeric environments. In this phylogenetic study we comprehensively sample the genera of Nyctaginaceae, with the following goals: (1) to evaluate the existing classification of Bittrich and Kühn (1993) , (2) to understand the biogeographic history of the family, and (3) to have a basis for understanding the evolutionary history of characters of historical taxonomic importance and the potential adaptive significance as manifested in their "tendency" to evolve repeatedly in lineages occurring in the deserts of North America.

MATERIALS AND METHODS

Sampling
Fifty-one species representing 25 genera of Nyctaginaceae were sampled. Taxa, voucher information, and GenBank numbers are given in Appendix 1. Our sampling is nearly comprehensive at the generic level, with representative species of every genus except Neeopsis, Cephalotomandra, Grajalesia, Cuscatlania, Boldoa, and Cryptocarpus. The genera omitted are monotypic, rarely collected, and/or of dubious distinction. For example, Boldoa purpurascens is often included in Salpianthus (Pool, 2001 ). All tribes and subtribes recognized by Bittrich and Kühn (1993) are included. Because different taxa have been found to be sister to Nyctaginaceae (Rettig et al., 1992 ; Behnke, 1997 ; Downie et al., 1997 ; Cuenoud et al., 2002 ), outgroups were selected from both Phytolaccaceae and Sarcobataceae. More distantly related taxa in the "core Caryophyllales," i.e., Aizoaceae, Molluginaceae, and Stegnospermataceae (Cuenoud et al., 2002 ), were also included to enable us to test the monophyly of Nyctaginaceae and to identify which taxa are sister to the family. For four species, data were obtained from two different accessions, and for two, GenBank sequences were used for some loci. "Phytolacca" is a composite of one GenBank sequence from P. acinosa and three new sequences from P. americana.

Molecular data
Genomic DNA was extracted from fresh, silica-dried, or air-dried (herbarium) leaf tissue using either Qiagen DNAeasy Plant Mini Kits or a modified CTAB method (Doyle and Doyle, 1987 ). Internal transcribed spacer (ITS) sequences were obtained using primers ITS4 and ITS5a (White et al., 1990 ; Stanford et al., 2000 ), which amplifies ITS1, 5.8S, and ITS2. Chloroplast ndhF sequences were obtained as two overlapping fragments using primers Nyct-ndhF1, ndhF972, Nyct-ndhF13R, and Nyct-ndhF22R. With the exception of ndhF972 (Olmstead and Sweere, 1994 ), these were designed based on GenBank ndhF sequences for Nyctaginaceae and Phytolaccaceae. Many samples, especially those from herbarium materials, were recalcitrant to PCR of long (>1 kb) fragments due to DNA degradation; for these, four additional primers (Nyct-ndhF6F, Nyct-ndhF8R, Nyct-ndhF13F, and Nyct-ndhF16R) were designed, based on sequences for Nyctaginaceae and Phytolaccaceae, and used in conjunction with the aforementioned primers, so that the gene was amplified in four overlapping fragments. The chloroplast intron rps16 was amplified using primers rpsF and rps2R (Oxelman et al., 1997 ), and rpl16 was obtained using primers F71 and R1661 (Jordan et al., 1996 ). Primer sequences and references are given in Table 2. PCR products were cleaned with Qiaquick columns (Qiagen, Valencia, California, USA). Cycle sequencing was performed using the BigDye Terminator v3.1 Cycle Sequencing Kit, and sequences were determined with an ABI 3700 DNA Analyzer (Applied Biosystems, Foster City, California, USA) in the Genetic Analysis facility in the Department of Biology at Duke University. Raw chromatograms were edited and assembled in Sequencher 4.1 (Gene Codes Corp., Ann Arbor, Michigan, USA). Sequence alignment was performed either by eye (ndhF) or in ClustalX (Thompson et al., 1997 ) (other regions) followed by manual adjustment in Se-Al (Rambaut, 1996 ). Across the entire data set, ITS1 and ITS2 were too variable to be confidently aligned, although the 5.8S region was highly conserved. Ambiguously aligned regions were excluded from further analyses of the entire data set, though they were used in analyses of more restricted taxon sets (see Restricted analyses).

Data analysis
Initial maximum parsimony (MP), maximum likelihood (ML), and Bayesian analyses were performed for each of the four loci. The 5.8S, not surprisingly, had low variation and produced poorly resolved trees; however, examination of the support values for the topology favored by each locus revealed no supported nodes in conflict. Therefore, the data sets were combined for further analyses.

MP analysis was performed using PAUP* version 4.0b10 (Swofford, 2002 ). A heuristic search was performed, with 1000 replicates of 10 random-addition sequences, tree-bisection-reconnection (TBR) branch swapping, MAXTREES set to autoincrease, MULTREES = yes. Support was evaluated using 1000 bootstrap replicates of 10 random addition sequences, TBR branch swapping, MULTREES = YES.

For the ML analysis, the data set was first examined using ModelTest 2.0 (Posada and Crandall, 1998 ), which selected a complex model of evolution (GTR + I + ). Ten random-addition replicates (TBR, MAXTREES set to autoincrease, MULTREES = yes) were run in PAUP*. Maximum-likelihood bootstrap support values were obtained by 100 replicates of single random-addition sequences, TBR branch swapping, MULTREES = yes.

Bayesian analysis was performed using MrBayes 3.1 (Ronquist and Huelsenbeck, 2003 ). For exploring the effect of different models for different partitions of the data, best-fit models for each partition were estimated in MrModelTest (Nylander, 2004 ), which selects the best-fit model from those available in MrBayes. The partitions were as follows: 1, all loci together; 2, nuclear 5.8S; 3, all chloroplast loci; 4, rpl16; 5, rps16; 6, ndhF; and 7, 8, 9, first, second, and third positions of ndhF, respectively. The models selected by MrModelTest for each partition are given in Table 3. Bayesian searches were then performed on the entire data set using four partition/model combinations: "B1," single model for all partitions, (1); "B2," nuclear and chloroplast, (2 and 3); "B4," all loci, (2, 4, 5, and 6); and "B6," all loci with separate models for each codon position of ndhF (2, 4, 5, 7, 8, and 9). For each combination, we executed four independent runs of 1 x 10⁶ generations each, sampling every 100th tree. After discarding trees from the burn-in (determined by visualizing the plateau in –lnL scores, approximately after 50 000 generations), we compared the posterior tree sets from each run by computing a 50% majority rule tree in PAUP*. No strongly supported topological differences (at posterior probability 95%) were found between the four runs of each model set. Therefore, the four posterior tree files for each set of models were combined into a single posterior tree file for purposes of assessing support values yielded by each set of models. These preliminary analyses were conducted including the partial ndhF sequence for Caribea; however, the B6 analysis was repeated without this sequence.

Restricted analyses
To gain resolution within and between closely related genera, our selection of loci encompassed a large range of sequence variation. Because both the ITS1 and ITS2 regions had to be excluded from the analysis of the complete data set due to questionable alignment (though the highly conserved 5.8S region was kept in the full matrix), following the analysis of the full data set, two restricted data sets were constructed to allow us to increase the number of included characters (Table 3) by reducing the taxon sampling to two distinct clades found in the full analysis. These restricted data sets comprised all included nucleotide positions in the full data set, plus sites that were unalignable across the breadth of taxa included in the full data set, but that were alignable within each of the restricted sets of taxa. The first restricted analysis group was comprised of North American herbs representing all taxa in the sister group to Allionia, whereas the second corresponded to the Pisonieae, Bougainvillea, Belemia, and Phaeoptilum (the "B&P" clade from the full analysis). The MP, ML, and corresponding bootstrap analyses were performed in the same fashion as in the full matrix and sensitivity analyses, with the exception that the ML models were reestimated in ModelTest.

Character data
The historical taxonomic significance given to pollen morphology and involucral bracts led us to examine these characters in a phylogenetic context. Pollen data follows the scheme of Nowicke, who identified four types in Nyctaginaceae (Nowicke, 1968 , 1970 , 1975 ; Nowicke and Luikart, 1971 ; Reyes-Salas and Martínez-Hernández, 1982 ; Chavez et al., 1998 ). Pollen type was coded as a multistate, unordered character. In many cases, the exact species included in our study were not examined in the published studies. If there was no indication of within-genus pollen polymorphism, that pollen type was assigned to all species in this analysis. However, multiple pollen types were recorded within Neea and Pisonia. Thus, only N. psychotrioides, which was examined by Nowicke, was coded unambiguously; other species of Neea and Pisonia were coded as polymorphic (states "1&3" and "1&4," respectively) to reflect this uncertainty in the assignment of ancestral states. The presence of involucral bracts was scored as present/absent. If only small subtending bracteoles occur (common in many taxa), this character was coded as "absent," mirroring the usage of this character in defining subtribe Nyctagininae. The occurrence of cleistogamous flowers was scored based primarily on literature sources (Spellenberg and Delson, 1974 ; Bittrich and Kühn, 1993 ; Levin, 2002 ; Spellenberg, 2003 ). Gypsophilic taxa were identified in literature sources (Waterfall, 1946 ; Parsons, 1976 ; Fowler and Turner, 1977 ; Turner, 1991 ; Harriman, 1999 ; Levin, 2002 ; Spellenberg, 2003 ; N. Douglas, personal observation). Taxa were identified as full gypsophiles (recorded only from gypseous soils), gypsum tolerant (recorded from both gypseous and nongypseous soils), or nongypsophilic. Taxa that do not occur in areas with gypsum outcrops were considered to be nongypsophilic. It is unlikely that transitions to or from full gypsophily could evolve with no intermediate gypsum-tolerant step; therefore, this character was analyzed as both unordered and ordered, with two steps required between nongypsophily and full gypsophily. Parsimony ancestral states of all characters were reconstructed with the program Mesquite 1.6 (Maddison and Maddison, 2006 ). Those terminals that were not assigned a single state, and branches that were not unambiguously resolved, are depicted as "equivocal."

RESULTS

Data matrix
The entire data matrix (Table 3) had a length of 5505 bp, of which 1771 were excluded due to ambiguous alignment, mainly due to the presence of length variation in ITS1 and ITS2 and in the two chloroplast introns, rpl16 and rps16. Of the remaining 3734 characters, 652 were parsimony informative.

Phylogenetic analysis of the complete dataset
The MP analysis resulted in 36 shortest trees (length: 2287, consistency index [CI]: 0.657, retention index: 0.809, rescaled CI: 0.531); however, the strict consensus (tree not shown) resolved all but two ingroup nodes. Thirty-nine nodes were supported with parsimony bootstrap values (MPBS) 70.

The best-fit model as determined by ModelTest (Table 3) using both a hiearchical liklihood ratio test (HLRT) and the Akaike information criterion (AIC) was a general-time-reversible model with a proportion of invariant sites and a gamma shape parameter (GTR+I+). The ML search returned a single ML tree, which was nearly identical to the MP topology, except in the placement of the genus Colignonia. This taxon is placed as sister to the large clade containing Acleisanthes and Boerhavia in the MP analysis (MPBS = 80) and is not resolved with strong support in any ML or Bayesian analysis. Overall, 38 nodes in the ML analysis were supported with likelihood bootstrap values (MLBS) 70.

Models determined by MrModelTest for each data partition in the Bayesian analyses are given in Table 3. On the basis of our preliminary examination of partitioned models, the signal in the data set apparently is strong, and the topology is not contingent on model selection: the tree topologies produced by the Bayesian B1, B2, B4, and B6 searches were consistent. The principal difference between them is in the level of support for the topology, with 37, 39, 40, and 40 nodes, respectively, supported by posterior probabilities (PP) 95%. Deletion of Caribea led to the resolution, with support of two additional nodes in the repeated B6 search, for a total of 42 nodes supported at greater than 95% PP. The topology of this Bayesian B6 consensus tree is identical to the ML tree. All further Bayesian support values refer to the B6 analysis.

The Nyctaginaceae are supported as monophyletic by ML (MLBS = 71) and Bayesian (PP = 100) analyses (Fig. 1). Interestingly, in the MP bootstrap analysis of this matrix, the monophyly of the Nyctaginaceae is not supported. Despite the inclusion of several outgroups, no single sister lineage emerges with strong support.

View larger version (24K): [in this window] [in a new window]   Fig. 1. Maximum-likelihood (ML) topology from the analysis of the entire data set. Parsimony bootstrap/ML bootstrap support values above branches, Bayesian posterior probability from the "B6" analysis below branches, "-" indicates bootstrap support value <50. tribes of Nyctaginaceae according to Bittrich and Kühn (1993) are in bold. "-" before unbold name signifies a subtribe of tribe Nyctagineae

View larger version (42K): [in this window] [in a new window]   Fig. 2. Phylogram of the maximum-likelihood topology from Fig. 1. Major clades referred to in text are highlighted

The position of Colignonia is not resolved in the ML and Bayesian analyses. The ML analysis resolves Colignonia sister to the B&P and NAX clades but with weak support. A position sister to only the NAX clade is supported in the MP analysis.

Sensitivity analyses
The deletion of taxa with significant missing data resulted in a matrix of 39 taxa with only 3.1% missing data, as compared to 58 taxa with 17.7% missing data in the full analysis (See Appendix 1). The MP/ML analyses of this matrix yielded trees (not shown) that had no well-supported nodes conflicting with the topology of the tree from the full matrix. The support for the monophyly of the Nyctaginaceae increased to 94/95 MPBS/MLBS, from –/71 in the analysis of the full data set. The high level of support found in this analysis for the monophyly of Nyctaginaceae indicates that the inclusion of many outgroups in the full matrix, including the quite distant Stegnosperma, may have affected the level of support in the MP analysis. Alternatively, high levels of missing data in the full data set may be responsible for low support values at this key node. Support for the placement of Cyphomeris decreased to 70/66 relative to the full analysis. Commicarpus and Allionia increased to 73/67 and 87/77, respectively; these nodes had not received strong support in any analysis of the full data set. The remainder of the comparable nodes were similarly supported between the full and sensitivity analyses.

Restricted analyses
For the two restricted analysis groups, 122 and 76 additional informative characters were gained with the inclusion of ITS1 and ITS2, respectively (Table 3). A small number of additional sites were gained from the chloroplast introns rps16 and rpl16 (<5 characters in either data set). ModelTest 3.7 selected a GTR+I+ model for each (Table 3) data set. For the first group (all taxa in the sister group to Allionia), MP and ML analyses produced a tree (Fig. 3) with improved resolution in the Anulocaulis + Nyctaginia clade. Though the placement of A. annulatus differs between the full matrix topology and the restricted analysis, the monophyly of Anulocaulis was well supported with bootstrap values of 72/89 MPBS/MLBS. Similarly, the full analysis resolves Okenia within a paraphyletic Boerhavia with low MP and ML bootstrap support, but 97% Bayesian posterior probability, yet the restricted analysis found Boerhavia strongly supported as monophyletic and sister to Okenia with high support (100/100). Boerhavia consists of two clades, corresponding to annual and perennial species, that were also found in the full analysis.

View larger version (46K): [in this window] [in a new window]   Fig. 3. Phylogram of the maximum-likelihood (ML) topology from the first restricted analysis. MP bootstrap/ML bootstrap support values are shown. Anulocaulis and Boerhavia are each supported as monophyletic

Character reconstructions
For each character reconstructed (Fig. 4), multiple state transitions are inferred. Tricolpate-spinulose pollen (Fig. 4a) appears to be the ancestral condition in the group, transitioning to a pantoporate-spinulose condition subsequent to the divergence of Salpianthus from the main lineage. The latter condition is found in nearly all members of the NAX clade, yet appears to predate that group. At least eight transitions among the four pollen types have occurred in the Nyctaginaceae. Considering the small number of Neea and Pisonia examined and the polymorphism exhibited by these genera, the number of transitions could be higher. Reconstruction of involucral bracts shows five gain/loss steps. This character is fixed within genera, thus this interpretation is likely to be affected only by the future inclusion of the remaining genera in the family. Only the inclusion of Cuscatlania, which has an involucre, could conceivably change the number of steps required. Cleistogamous flowers are uniquely derived in four genera. Gypsophily requires nine or 13 steps to explain, depending on whether it is considered to be an unordered or an ordered character. Reconstructions were performed only on the ML topology from the full analysis. Adjusting the positions of Okenia and Nyctaginia to reflect the topology from the restricted analysis (Fig. 3) results in the branches leading to Nyctaginia + Anulocaulis and Nyctaginia + Anulocaulis + Okenia + Boerhavia being resolved as nongypsophilic. Treating gypsophily as an unordered character has the same result. Otherwise, the alternative topology has no substantive effect on the conclusions we make regarding the degree of homoplasy shown by the remaining three characters shown in Fig. 4.

View larger version (58K): [in this window] [in a new window]   Fig. 4. Parsimony reconstruction of (A) pollen morphology, (B) involucre presence, (C) cleistogamous flowers, and (D) gypsophilic habit (based on ordered characters). See text for sources of characters used in reconstructions

Phylogeny of Nyctaginaceae
The earliest branching lineage in Nyctaginaceae, the Leucastereae (Fig. 1), had been previously recognized as a natural group on the basis of arborescence, a stellate indumentum, and tricolpate pollen (Heimerl, 1934 ; Bittrich and Kühn, 1993 ). The Boldoeae, an herbaceous group native from the Galapagos to northwestern Mexico and the Caribbean, are represented in this study by Salpianthus. These two lineages had been predicted to be basal or outside of Nyctaginaceae on the basis of apparent pleisomorphies such as alternate leaves and bisexual flowers (Bittrich and Kühn, 1993 ). The anthocarp structure is absent in Leucastereae and Boldoeae, although the unexpanded perianth does persist around the fruit. Persistent tepals are also found in many Phytolaccaceae. However, the perianth consists of free tepals in most Phytolaccaceae and all of subfamily Rivinoideae (except Hilleria, in which three of four tepals are partially fused, (Rohwer, 1993 )). In Nyctaginaceae, including Leucastereae and Boldoeae, tepals are fully connate.

Within the B&P clade (Fig. 2), Phaeoptilum is found to be sister to Belemia, rendering the Bougainvilleeae paraphyletic. The Pisonieae are found to be sister to Pisoniella, which had been included in that tribe by Heimerl (1934) but was removed to subtribe Colignoniinae by Bittrich and Kühn (1993) following the suggestion of Bohlin (1988) . The reasoning behind this move is mysterious, and in light of our results, it appears to have been unwarranted. Pisoniella possesses a straight embryo like other Pisonieae, and the large coriaceous anthocarps are provided with viscid glands along the ribs, much like those in Pisonia (Heimerl, 1934 ).

Within Pisonieae, Neea and Guapira form a clade (Fig. 2). These genera are distinguished primarily by whether the stamens are included (Neea) or exserted (Guapira). Our sampling is extremely limited in these two large genera, with only five accessions to represent ca. 150 species, though we were able to include accessions from geographically disparate locales. Neither genus forms a monophyletic group. This conclusion has been occasionally anticipated (e.g., Pool, 2001 ). It is unclear whether our sampling simply happened to include misclassified species in otherwise good genera, or whether this paraphyly is representative of Neea and Guapira generally. Much more intensive sampling is clearly needed to understand the relationships of the species in these genera, and it would be imprudent to attempt to reclassify them until a more detailed study is made including phylogenetic, morphological, and distributional data. Unfortunately, collections of these dioecious trees often do not include individuals of both sexes. Also, the tendency of many Pisonieae to oxidize when dried has left many descriptions lacking crucial information concerning the color of fruits. Therefore, the taxonomic literature is quite confused and species limits are known not much better than when Standley (1931a , p. 73) wrote that, "I know of few groups of plants in which specific differences are so unstable and so baffling ... particularly in Neea, Torrubia [= Guapira] and Mirablis, no single character seems to be constant." Finally, in this study we did not attempt to infer the ages of lineages, yet it appears that the branch lengths in the Neea + Guapira clade are comparatively short, especially considering that this clade can be expected to accommodate as many as 150 species (Fig. 2). A similar pattern has been noted in other radiations of neotropical trees, e.g., Inga (Fabaceae) (Richardson et al., 2001 ). If the pattern of relatively short branches inferred between species was upheld with the inclusion of a larger sample of taxa and more rapidly evolving markers, it would point to this clade as another example of rapid diversification in the neotropics.

Tribe Nyctagineae is broadly paraphyletic. As mentioned, Pisoniella and Phaeoptilum are not found in this study to be closest relatives of any other Nyctagineae. Based on pollen morphology, Bohlin (1988) has suggested that Colignonia (subtribe Colignoniinae) has affinities to the tribe Mirabileae of Heimerl (1934) , which roughly corresponds to the tribe Nyctagineae and the NAX clade. Colignonia may in fact be sister to the NAX clade as suggested by the MP analysis or to the NAX + B&P clade as suggested by the ML analysis (Fig. 1). Tribe Nyctagineae also does not include Abronia or Tripterocalyx (tribe Abronieae, Fig. 1). There are certain characters of the Abronieae that are anomalous within the Nyctagineae (and the NAX clade) and that justified recognition at a higher taxonomic level, namely, tricolpate pollen and linear stigmas. The two genera in the tribe have long been thought to be a natural group and are often synonymized (Heimerl, 1934 ; Bittrich and Kühn, 1993 ), though most authors have maintained the two genera (Galloway, 1975 ; Spellenberg, 2003 ). The Abronia + Tripterocalyx clade is characterized by the combination of an umbellate inflorescence of salverform flowers with included stamens and style, an involucre, anthocarps with typically well-developed wings or lobes, and a mature embryo with a single cotyledon.

Anulocaulis and Nyctaginia are classified in different subtribes in the classification of Bittrich and Kühn (1993) , presumably based on the presence of an involucre in Nyctaginia. Both genera are succulent perennial herbs, and the turbinate fruits with umbonate apices of Nyctaginia capitata strongly resemble those of Anulocaulis eriosolenus. They differ in many characters, including flower color (red-orange in Nyctaginia vs. white to pink in Anulocaulis) and flowering time (flowers of Nyctaginia are open during the day, while in Anulocaulis anthesis is at sunset or later and flowers wilt in the morning). While the full matrix ML tree (Fig. 1) indicates that Anulocaulis may not be monophyletic, this relationship is poorly supported (MPBS/MLBS/PP = 64/55/63). In the restricted MP and ML analysis (Fig. 3), however, a monophyletic Anulocaulis is more strongly supported (MPBS/MLBS = 72/89). Therefore, we see no compelling reason to question the taxonomic status of Anulocaulis.

Anulocaulis + Nyctaginia are sister to a strongly supported clade containing Boerhavia and Okenia. Like the previous instance, Okenia resolves within Boerhavia in the full matrix ML topology (Fig. 1), but support for this relationship is only moderately significant in the Bayesian analysis of the full data set (PP = 97) and weakly supported by MPBS and MLBS (67/69). Conversely, Boerhavia is strongly supported as a monophyletic group in the MP and ML analyses of the restricted data set (MPBS/MLBS = 100/100, Fig. 3). Vegetatively, Okenia strongly resembles most Boerhavia in its decumbent habit, and subequal opposite leaves with sinuate or undulate margins. The flowers of Okenia, though larger, are similar in color to some perennial Boerhavia from the Chihuahuan Desert. Finally, Okenia is annual, a condition found in one clade of Boerhavia. However, Okenia is strikingly different than Boerhavia in its unique reproductive biology: it produces aerial flowers, but the large, spongy fruits are geocarpic, with peduncles elongating greatly after fertilization and the fruits maturing several centimeters belowground. The relationship between these two genera is deserving of more study.

Biogeographical patterns
The basal lineages of Nyctaginaceae (Boldoeae, Leucastereae, Colignonia, Bougainvilleeae, and Pisonieae [including Pisoniella]) are fundamentally South American. Though some taxa have representatives or populations in (sub)tropical North America, (Salpianthus, Neea, Guapira, Pisonia, Pisoniella), their distributions all include the neotropics, and phylogenetically they are interspersed with neotropical endemics. The widespread tropical genus Pisonia possesses extremely viscid anthocarps, which aid dispersal, frequently by seabirds (Burger, 2005 ). The sole genus not native to the Americas is Phaeoptilum, endemic to arid southwestern Africa. This monospecific genus is closely related to Belemia and Bougainvillea, both from eastern and southern South America. Phaeoptilum is morphologically quite distinct from its sister taxon Belemia, though vegetatively it resembles the xeric-adapted Bougainvillea spinosa. The early Cretaceous date (130–90 Ma) for the opening of the south Atlantic (Smith et al., 1994 ) makes vicariance an unlikely explanation for this disjunction. Dispersal seems more likely, and while there is no specialized dispersal structure on the anthocarp of Belemia, both Bougainvillea and Phaeoptilum have compelling (albeit different) adaptations for wind dispersal. Phaeoptilum produces winged anthocarps highly similar to those found in Tripterocalyx and some species of Acleisanthes. In Bougainvillea, most species display three showy bracts, each fused to a solitary flower. In fruit each involucral bract remains fused to a fruit and acts as a wing, the structure functioning as a unit of dispersal (Ridley, 1930 ).

The North American Xerophytic Clade has diversified in the deserts of the southwestern United States and northwestern Mexico. Every genus is confined to or has representatives in this region. Widespread taxa in this clade, namely Commicarpus and Boerhavia, possess glandular fruits, which have most likely aided bird-dispersal in a manner similar to that of Pisonia. Two red-flowered Boerhavia, B. coccinea and the similar B. diffusa are widespread in most tropical and subtropical areas. Boerhavia diffusa appears to have naturally dispersed from the Americas, though the confused taxonomy of this species and B. coccinea in regional floras makes this difficult to evaluate, and both of these species are frequently transported by human activity. The "repens" complex in Boerhavia (B. repens and related species) is widespread in coastal habitats throughout the tropical Pacific and Indian oceans to the Arabian Peninsula, along with B. dominii from Australia. Like the red-flowered perennial Boerhavia mentioned, these species also have viscid glandular anthocarps. Okenia is found in deep sand dune habitat along the Pacific and Caribbean coasts of Mexico and Central America, with a disjunct population in southern Florida. Other authors (Heimerl, 1934 ; Fowler and Turner, 1977 ; Thulin, 1994 ; Levin, 2002 ; Spellenberg and Poole, 2003 ) have discussed the remarkable disjunctions of Acleisanthes somaliensis and Mirabilis himalaicus from east Africa and southern Asia, respectively. These appear to be attributable to long-distance dispersal events, due to their derived position within otherwise exclusively American clades (Levin, 2000 ; N. Douglas, unpublished data).

Pollen and involucre evolution
Tribal and subtribal classifications (Table 1) of the Nyctaginaceae have relied heavily on a few characters, such as pollen morphology and the development of an involucre. However, divisions based on these characters are not supported by our results because these characters have a high degree of homoplasy among genera.

Parsimony reconstruction of pollen type across Nyctaginaceae (Fig. 4a) shows that substantial homoplasy exists (11 changes), involving three of the four types diagnosed by Nowicke (Nowicke, 1970 , 1975 ; Nowicke and Luikart, 1971 ). Pantocolpate grains may constitute a synapomorphy for Belemia + Phaeoptilum. It has been noted that large, desiccation-resistant, pantoporate pollen grains, equipped with pore plates, were found primarily in the herbaceous desert taxa (Nowicke and Luikart, 1971 ). Specific correlations between large and/or polyaperturate grains and habitat in angiosperms have not been adequately investigated. In a study of ecological correlates of pollen morphology in a wide selection of angiosperms (Lee, 1978 ), there was an extremely weak correlation of pore number with width and with "temperature." According to our reconstructions, the origin of pantoporate-spinulose pollen predates the major radiation of desert taxa in the NAX clade. However, Colignonia and Pisoniella have much smaller grains than do the remaining taxa with pantoporate-spinulose pollen (Colignonia = 25–35 µM, Pisoniella = 30–37 µM, Nowicke and Luikart, 1971 ; N. Douglas, unpublished data). Therefore, it would seem best to consider grain size as a variable separate from grain shape and exine structure.

Within Nyctagineae, the subtribes Nyctagininae and Boerhaviinae were separated by the presence or absence of an involucre subtending the inflorescence. In subtribe Nyctagininae, the involucre of Mirabilis is comprised of fused bracts; the remaining genera possess involucres of distinct bracts. The involucre in Bougainvillea is distinctive; fruits of Bougainvillea retain a large involucral bract as discussed. Involucres have no known dispersal function in any of the other taxa; they likely serve merely to protect the flower buds and developing fruits or discourage nectar-robbing insects (Cruden, 1970 ). Parsimony reconstruction of this character on the molecular topology (Fig. 4b) indicates that, for involucres, there are at least five gain/loss steps in the family, four in the NAX clade, which contains the members of the Nyctagineae-Nyctagininae, Nyctagineae-Boerhaviinae, and Abronieae, reflecting the artificial nature of this classification. In this analysis, the character was treated in a very simplistic fashion, reflecting nothing more than taxonomic convention. Comparative developmental studies may shed light on deeper homologies or convergences, especially as they relate to the subtending bracts found in many genera. The selective benefits involved in the expression of this structure could be revealed by appropriate ecological investigations.

Self-compatibility and cleistogamy
The production of obligately selfing flowers is obviously contingent on the ability of plants to self-pollinate and produce fertile progeny. Our incomplete knowledge of reproductive systems in Nyctaginaceae means that an unambiguous reconstruction of self-compatibility is not currently possible. However, several studies have addressed mating systems in select Nyctaginaceae: sporophytic self-incompatibility (SI) is known in Bougainvillea (Zadoo et al., 1975 ; López and Galetto, 2002 ). Some Mirabilis (sect. Quamoclidion) and Abronia macrocarpa fail to set seed when self-pollinated (Cruden, 1973 ; Williamson et al., 1994 ), but the basis for incompatibility is not known in these genera. The Pisonieae are usually dioecious and are thus self-incompatible, although in these genera there are occasional monoecious or hermaphroditic species (e.g., Pisonia brunoniana) for which the mating system has not been studied (Sykes, 1987 ). Evidence suggests that many genera in the NAX clade are self-compatible: in addition to the production of cleistogamous flowers in four genera, Boerhavia and some Mirabilis are known to have a delayed self-pollination mechanism whereby the style curls and encounters the anthers as the flower wilts (Chaturvedi, 1989 ; Hernández, 1990 ; Spellenberg, 2000 ). Finally, flowers protected from pollinators have set viable seed in Abronia umbellata Lam. (McGlaughlin et al., 2002 ) and Colignonia (Bohlin, 1988 ).

Reasoning from these data, we can make certain inferences regarding the evolution of mating systems in Nyctaginaceae. Explanations for current distribution of mating systems family must incorporate one, or some combination of both, of the following scenarios. Which one is preferred depends on the likelihood of self-compatible lineages giving rise to lineages with an inability to self-fertilize, and the implications of either scenario are interesting.

One scenario, and the most parsimonious given our current knowledge, is that there have been at least three independent derivations of SI from a self-compatible ancestor. A single change can account for the Pisonieae and Bougainvillea, one for the derived Mirabilis sect. Quamoclidion, and one for Abronia macrocarpa. It is often assumed that outcrossing species are not derived from selfing ancestors and that selfing lineages are an evolutionary "dead end" (Fisher, 1941 ; Stebbins, 1974 ; Lande and Schemske, 1985 ). In the case of Nyctaginaceae, however, the question is whether it is possible that self-incompatible species have arisen from self-compatible ancestors. It would seem that populations making this transition would be subject to most of the forces that affect the balance of selfing and outcrossing in self-compatible populations. A recent study of s-locus polymorphism in Solanaceae (Igic et al., 2006 ) has shown that losses of SI are irreversible in that family. The "cost" of developing the complex genetic systems necessary for SI would be added to the transmission advantage of alleles promoting self-fertilization (Uyenoyama et al., 1993 ); these factors must count against a hypothesis of multiple transitions to SI in one family.

Conversely, if we assume that SI is ancestral and has been lost repeatedly, transitions from SI to self-compatibility have occurred a minimum of six times (in Colignonia, Acleisanthes, some Abronia, two or more times in Mirabilis, and finally in the clade sister to Mirabilis). This represents a doubling of the number of evolutionary steps required to explain the distribution of known Nyctaginaceae mating systems. Other authors have discussed the merits of parsimony weighting schemes or maximum-likelihood approaches to testing the irreversibility of selfing (Barrett et al., 1996 ; Bena et al., 1998 ; Takebayashi and Morrell, 2001 ). In these cases, it may not be possible to escape a circular argument employing only phylogenetic evidence, because a weighting scheme favoring losses of SI assumes the conclusion. In Solanaceae (Igic et al., 2006 ), evidence of ancient polymorphism at the incompatibility locus itself was required to demonstrate the irreversibility of the loss of SI. In our case, the most convincing resolution will come when SI is characterized in Mirabilis sect. Quamoclidion and SI Abronia. If in these taxa and any others that may yet be discovered to be self-incompatible the genetic basis for SI can be identified, homology could be assessed and the ancestral functionality of the underlying mechanism could be tested.

Assuming the derived state is self-compatibility, of these six lineages, three have given rise to cleistogamous/chasmogamous lineages, and four gains of cleistogamy are required to explain the distribution of the character in Nyctaginaceae (Fig. 4c). Interestingly, the cleistogamous genera are all perennial, which should be less susceptible to selection pressure for reproductive assurance than annuals (Barrett et al., 1996 ). Alternatively, cleistogamous flowers can function to maximize seed set when resources, rather than pollinators, are limiting (Schemske, 1978 ). These hypotheses are both applicable to the cleistogamous Nyctaginaceae, though distinguishing between them may be difficult, because pollinators in desert environments tend to be scarce when water is scarce. Spellenberg and Delson (1974) found that Acleisanthes (Ammocodon) chenopodioides, with a generalized flower morphology and a diurnal pollinator fauna, produced roughly equal numbers of seeds from cleistogamous and chasmogamous flowers, and did not have a strong seasonal pattern in the production of cleistogamous flowers. In contrast, Acleisanthes longiflora, a species with large, specialized hawkmoth-pollinated flowers, produced the majority of a season's seeds from cleistogamous flowers produced preferentially in the dry early summer when sphingid moths are less active. This may suggest that cleistogamy in this genus is insurance against reproductive failure due to the absence of pollinators in some years.

Gypsophily
Parsimony reconstruction of gypsophily in Nyctaginaceae (Fig. 4d) indicates that gypsophiles and gypsum-tolerant species are widely dispersed in the NAX clade. With the current sampling, the ancestor of this clade is inferred to be nongypsophilic (whether or not the character is considered "ordered"), indicating that gypsum tolerance is derived multiple times. This conclusion is tenuous for two reasons. First, gypsum outcrops are common in the Chihuahuan Desert but less so in other parts of the ranges of the NAX genera. We are unable to rule out the possibility that taxa coded in this analysis as "nongypsophilic" are actually gypsum-tolerant, but simply do not occur in areas with gypsum soils.

Second, there are two Mirabilis [M. nesomii Turner and M. linearis (Pursh) Heimerl] which are gypsophilic (Turner, 1991 ) and gypsum-tolerant (R. Spellenberg, New Mexico State University, personal communication), respectively. These species, both in section Oxybaphus, are close relatives of the oxybaphoid M. albida, a nongypsophile included in this study. It is possible to add gypsophilic taxa as sisters to M. albida on our topology, so that the resolution of the ancestor of the NAX clade becomes equivocal, with ACCTRAN reconstruction as gypsum-tolerant, and DELTRAN as nongypsophilic. The same reconstruction would be made for the ancestors of Commicarpus and Abronia + Tripterocalyx. The sensitivity of the reconstruction at these key nodes to sampling artifacts indicates that in order to reconstruct the history of gypsophily in this clade, it will be necessary to undertake more intensive phylogenetic sampling at the species level, investigating an appropriate sample of nongypsophilic taxa closely related to known gypsophiles.

Even if we cannot know the gypsum tolerance of the ancestor of the NAX clade based on existing data, it is evident that there are at least four instances of strong gypsophily evolving in the family. It would be profitable to investigate the ecology of these gypsophytes and their relatives in the NAX clade. An experimental approach investigating whether or not seedlings of nongypsphiles have the latent ability to establish on gypseous crusts would disentangle the expression of gypsum tolerance from biogeographic complications, clarify the phylogenetic distribution of gypsum tolerance and perhaps reveal the nature of the adaptation(s) involved.

It is possible that establishment on gypsum is facilitated by some sort of modification to the radicle. Alternatively, because germination in a desert environment is always risky, adaptations to gypsum soils may differ little from germination strategies of desert taxa generally. Possible strategies could serve to optimize the timing of germination, minimize the risk of all seedlings perishing or increase the length of time a seedling has to establish itself. These could include high germination rate at low temperatures and various forms of bet-hedging, such as seed heteromorphism and variable seed dormancy (Escudero et al., 1997 ). The production of mucilage upon wetting by the seed coat presumably increases the local availability of water and upon drying, anchors the seed (Romao and Escudero, 2005 ). Some of these traits are known in Nyctaginaceae. For instance, production of mucilage by the anthocarp is common in both gypsophilic and nongypsophilic taxa in the NAX clade (Spellenberg, 2003 ), and fruit/seed heteromorphism is known in Abronia and Tripterocalyx (Wilson, 1974 ).

Understanding when in their history Nyctaginaceae became gypsum-tolerant will clarify whether homoplasy is best explained by answering the question "how do species become gypsum-tolerant?" or "why are certain species found only on gypsum?" If it turned out that gypsum tolerance was ancestral in the NAX clade, then experiments may reveal the reasons full gypsophiles do not occur on more typical soils.

The tendency of Nyctaginaceae to evolve cleistogamy and gypsophily has been shown to the extent that we have demonstrated that the high level of homoplasy for these traits is restricted to the NAX clade. In neither case are we able to conclusively identify the largest group capable of evolving the trait. Largely due to the phylogenetic position of Acleisanthes (with gypsophilic, cleistogamous species), we infer that it is possible that the ancestor of the entire NAX clade was predisposed to evolve these traits. In the case of cleistogamy, the topology indicates either that SI mechanisms develop easily in Nyctaginaceae, or that once self-compatibility emerges, there is a high chance of cleistogamy following. If the latter situation is correct, the explanation for the large number of cleistogamous species in the NAX clade must ultimately rely on explaining the frequent loss of SI, though the proximate cause is more likely related to resource or pollinator limitation in xeric environments. With gypsophily, it remains to be seen what trait(s) allow for tolerance of gypsum soils and when they evolved and what factors act exclude to gypsophiles from nongypsum soils.

The present study is the first to provide a comprehensive genus-level examination of the phylogeny of Nyctaginaceae. Though sampling of Caribea, Cuscatlania, Cephalotomandra, Grajalesia, and Neeopsis would be desirable, the current level of sampling is sufficient to draw several useful conclusions with bearing on future studies of the family. Aside from providing a framework for future taxonomic revisions, it raises interesting evolutionary questions regarding biogeography, reproductive biology, and edaphic endemism. To a degree, this work may be considered a case study into the practical issues that may arise in an investigation of tendencies in character evolution. New insights will be gained with a combination of phylogenetic work at finer taxonomic scales and experimental data to better understand the natural history of individual species, especially those in the xerophytic clade.

APPENDIX 1.

Taxa, GenBank accession numbers, and voucher information used in this study. Regions not sampled are indicated by a dash. Cultivated plants were obtained from the following sources: