GONDWANAN VICARIANCE OR DISPERSAL IN THE TROPICS? THE BIOGEOGRAPHIC HISTORY OF THE TROPICAL MONOCOT FAMILY COSTACEAE (ZINGIBERALES)

Costaceae are a pantropical family, distinguished from other families within the order Zingiberales by their spiral phyllotaxy and showy labellum comprised of five fused staminodes. While the majority of Costaceae species are found in the neotropics, the pantropical distribution of the family as a whole could be due to a number of historical biogeographic scenarios, including continental-drift mediated vicariance and long-distance dispersal events. Here, the hypothesis of an ancient Gondwanan distribution followed by vicariance via continental drift as the leading cause of the current pantropical distribution of Costaceae is tested, using molecular dating of cladogenic events combined with phylogeny-based biogeographic analyses. Dispersal-Vicariance Analysis (DIVA) is used to determine ancestral distributions based upon the modern distribution of extant taxa in a phylogenetic context. Diversification ages within Costaceae are estimated using chloroplast DNA data (trnL–F and trnK) analyzed with a local clock procedure. In the absence of fossil evidence, the divergence time between Costaceae and Zingiberaceae, as estimated in an ordinal analysis of Zingiberales, is used as the calibration point for converting relative to absolute ages. The results of the temporal analysis based on extant taxa indicate that the initial diversification within Costaceae occurred approximately 65 million years ago, long after the final break up of the Gondwanan supercontinent. Considering this minimum age of diversification, potential scenarios for the current biogeographic patterns found in Costaceae are presented in a temporal and spatial context. The evolution of specialized floral forms associated with specific pollinators is also discussed within the biogeographic framework.


INTRODUCTION
Pantropical distributions of plants and animals that are lacking intrinsic mechanisms for long-distance dispersal are often considered to indicate historic Gondwanan distributions, with the break-up of the supercontinent leading to vicariance-mediated diversification within lineages (Baum et al. 1998;Hurr et al. 1999;Bargelloni et al. 2000;Trewick 2000).While it has been suggested that the pantropical Costaceae exhibit a Gondwanan distribution pattern, this assumption has never been tested by cladistic or historical biogeographic methods.Recent studies of the monocotyledonous orders and families therein (Bremer 2000(Bremer , 2002;;Vinnersten and Bremer 2001) show that the origin of the major monocotyledonous lineages date back only to the early Cretaceous, approximately 100 million years ago (mya).This is about the time of the separation of major portions of Gondwana, notably the final separation of Africa from South America around 110-90 mya (Metcalf 2002;Sanmartı ´n and Ronquist 2004) and thus vicariance via continental drift could have provided a mechanism for diversification within the major monocot lineages.Based on recent work dating the origin of and major diversification events within the monocot order Zingiberales (Kress and Specht 2006), it is quite possible that Zingiberales date back to 110 mya, making it possible that late Cretaceous continental drift played a major role in the diversification of basal lineages of the order.
One of the four ''Ginger families,'' Costaceae form part of the crown group of Zingiberales.The four families in this crown clade taken together have a modern pantropical distribution, however the ancestral distribution of the ginger families clade is optimized as neotropical (Kress and Specht 2006).Modern Costaceae are pantropically distributed with their center of species diversity located in South and Central America (ca. 95 neotropical species,ca. 15 African,ca. 23 Southeast Asian).According to previous analyses (Kress and Specht 2006), the common ancestor of Costaceae and its sister family, Zingiberaceae, was distributed in Africa, America, Melanesia and Southeast Asia.This geographic area is considered consistent with an historic Gondwanan distribution pattern.
Costaceae have small (2-4 mm long) arilate seeds contained in a thin-walled capsulate fruit that can be dry and dehiscent or fleshy and indehiscent, in the latter case the seeds being released with the decaying of the fruit wall.Little is known about seed dispersal, although ants and birds are thought to be involved in some species (Maas 1972;Schemske 1983), with the potential for bird dispersal of seeds effectively increasing the changes of long distance colonization events across terrestrial barriers and open water.Floating capsules could potentially be involved in dispersal across long distances in water, although the longevity of the seeds under such circumstances has not been tested.
Previous work at the ordinal level (Kress and Specht GALLEY 632 2006) shows Costaceae diverged from Zingiberaceae approximately 105 mya.At this time, Gondwana was still largely intact.It is possible, then, based on biogeographic and temporal analyses, that the break-up of Gondwana influenced speciation within the Costaceae lineage during the mid-late Cretaceous, and long-distance dispersal and colonization events would not be required to explain the current wide distribution in modern-day wet tropical forests.Taxon sampling within the family is not sufficient in the ordinallevel study to date major cladogenic events, thus the role of continental vicariance or long-distance dispersal in speciation and diversification within Costaceae cannot be determined.
The following is a presentation of a more detailed investigation into the rates and dates of diversification within the Costaceae lineage, with special emphasis on the historical biogeography of the clade.Utilizing a chloroplast DNA-derived molecular phylogeny for a broad sampling of taxa representing the full taxonomic and biogeographic diversity of the family, the historical patterns of biogeography and dates of diversification are analyzed to test the Gondwanan-origin and dispersal/vicariance scenario that could best explain the events leading to the current pantropical distribution of the group.Biogeographic scenarios for species radiations within the family are also discussed, and associations with specific pollinators as a potential factor in the increased species diversity in the neotropics is addressed.

Phylogenetic Reconstruction
A data matrix was constructed with chloroplast DNA sequence data (trnL-F and trnK, including the full matK coding region) for 36 ingroup taxa (out of ca.125 total Costaceae) and 1 outgroup taxon (Siphonochilus J. M. Wood & Franks, Zingiberaceae) [Table 1].Ingroup taxa represented all major monophyletic lineages within Costaceae (Specht in press) and were chosen to reflect the full geographic distribution of the family, with multiple samples within each area so as to enable dating of nodes that are common to unique geographic regions (see Table 1).The complete chloroplast DNA data set yielded a total of 4066 unambiguously aligned characters (811 informative) for 37 taxa with no missing data except for gapped regions treated as missing data.Parsimony analysis was performed using PAUP* vers.4.0b10 (Swofford 2003) as a heuristic search with tree bisection reconnection (TBR) branch swapping, characters unordered, and stepwise addition for matrix building with 100 replicates and one tree held at each step.Insertions/deletions (indels) were coded as unique characters (21 total) and included in the parsimony analysis for topology building, but both coded indels and gapped regions were excluded in the branch length estimates for the subsequent maximum likelihood analysis.

Model Selection and Branch-Length Estimation
Modeltest vers.3.0 (Posada and Crandall 1998) was used in combination with PAUP* vers.4.0b10 to compare likelihood models and select the simplest model for estimating branch lengths that is not significantly different from the best-fitting model, using the Akaike Information Criterion (AIC) and likelihood ratio tests (LRT) with the default alpha level of significance (0.01).The use of simpler models decreases the error of estimated parameters, including branch length estimations critical for molecular dating analyses (Swofford and Olsen 1990).A parsimony-derived starting tree was used to select the model and estimate parameters (search ϭ heuristic, addition sequence ϭ random, number of replicates ϭ 100, branch swapping ϭ TBR).A full heuristic maximum likelihood search was performed with PAUP* vers.4.0b10 using the appropriate model with base frequencies and gamma shape parameter estimated from the data, 10 random addition sequence replicates retaining all minimal trees and TBR branch swapping.Analyses were performed with and without the ''enforce molecular clock'' option selected and a LRT was performed using the log likelihood scores from these searches in order to test for adherence to a global molecular clock.

Divergence Time Estimation
A pairwise relative rates test was used to investigate rate heterogeneity among and between taxa, implemented in Hy-Phy vers.0.95 beta (Kosakovsky-Pond andMuse 1998-2003): model ϭ F81, local option and all parameters constrained, outgroup ϭ Siphonochilus decorus (Druten) Lock.This pairwise test examines rate constancy between the outgroup taxon (Siphonochilus) and two ingroup taxa simultaneously, testing for divergence from the expected pairwise rates under the null assumption of a molecular clock.The output provides information on statistical similarity of rates for each combination of taxa.For each pairwise comparison, taxa were grouped into bins that showed no statisticallysignificant dissimilarity from the expected substitutions per branch, as indicated by the absence of an asterisk (*) in the Hy-Phy output table.Taxa for which significantly dissimilar pairwise rates from the expected (as indicated with *, **, or *** as indications of statistically dissimilar rates at P ϭ 0.05, 0.01 and 0.001, respectively) are noted and are grouped with taxa with similar pairwise relative rates based on multiple independent pairwise comparisons.Each bin was analyzed by investigating the individual branch lengths leading to the terminal taxa included in the bin, and an individual rate class corresponding to each bin was assigned to the appropriate branch or branches determined to be the cause of the rate heterogeneity based on visual inspection of branch lengths on the unconstrained chloroplast DNA phylogram (Jordan et al. 2003).
As with the Zingiberales data (Kress and Specht 2006), the local molecular clock method (Yoder and Yang 2000;Yang and Yoder 2003) was used to analyze the molecular data and estimate node ages for the chloroplast DNA data (trnL-F and trnK sequence data for 37 ingroup taxa plus one outgroup taxon).The local clock option of PAML vers.3.14 (Yang 1997(Yang , 2003) ) was executed with four rate classes (0-3) assigned to designated branches on the tree.Using the Combined Analysis option in PAML (clock ϭ 3), the chloroplast sequence data was divided into four data partitions: 1 st , 2 nd , and 3 rd codon positions within the matK coding region of the trnK gene region (partitions 1-3), and non-coding for the chloroplast intron and intergenic spacer regions (partition 4).Different alpha and kappa values were obtained for each data partition, thus accommodating heterogeneity in the combined data set.The baseml program of PAML was run with the following options: model ϭ 2 (F81), Mgene ϭ 4 (different rate ratio and frequency parameters for different partitions with proportional branch lengths), estimate kappa, estimate alpha, Malpha ϭ 1 (different gamma distribution for different parameters), ncatG ϭ 5 (number of categories), clock ϭ 3 (Combined Analysis), method ϭ 0 (simultaneous).Option G was specified in the sequence file, which, in combination with the Mgene option, accounts for heterogeneity in the data.Standard errors were obtained as part of the algorithm.
At least one calibration point is required to convert rela-  (Kress andSpecht 2005, 2006).Based on the Zingiberales results, there are two potential dates that can be used for calibration: the age of the node connecting Costaceae and Zingiberaceae (i.e., the date of divergence, 105 mya) and the age associated with the initiation of cladogenesis within Costaceae (i.e., date of diversification, 52 mya).However, the 52 mya date is the result of a divergence estimate using only four exemplar taxa in the crown group Costaceae.This age does not take into consideration the divergence dates of the basalmost members of the family, nor does it consider any potential extinction events that would indicate an earlier date of diversification.For these reasons, the age estimation for diversification of Costaceae was calibrated with the Costaceae-Zingiberaceae divergence age of 105 Ϯ 4 mya, using 101 mya and 109 mya in separate analyses in order to bracket the calibration age according to standard error calculations from Kress and Specht (2006).

Biogeographic Analysis (DIVA)
The topology obtained from the chloroplast data set was analyzed with DIVA vers.1.1 (Ronquist 1996), with geographic areas scored as presence or absence characters for each of the 37 taxa.The areas of distribution used for the DIVA analysis were: tropical Americas (divided into Central America [CAM] and South America [AM]), Africa (AF), Southeast Asia (SE), and Melanesia (ML, New Guinea and northern Australia).These areas were chosen as representatives of the major geographic areas of endemism for Costaceae and because defined taxonomic divisions (species and clades) do not typically span more than one of these areas.Coding of wide distributions can complicate reconstruction of the basal nodes, leading to reconstructions that consist of all possible combinations without providing resolution necessary to determine past biogeographic patterns (Sanmartı ´n and Ronquist 2002).The outgroup was coded as present in Africa based on the current distribution of basal-most extant Zingiberaceae.The number of areas allowed at each node (maxareas) was restricted to two, three, and four in independent analyses in order to look for optimal solutions that restricted the number of optimal basal node distributions reconstructed without significantly increasing the number of required dispersal events (i.e., compromising parsimony as defined by DIVA algorithm: Ronquist 1996).

Phylogenetic Analysis
The resulting topology (Fig. 1) was fully resolved and is identical to the phylogenetic results of a combined molecular and morphological analysis for Costaceae (Specht, in press), with the exclusion of the two Paracostus-clade taxa, Costus paradoxus K. Schum.and C. englerianus K. Schum., of the Asian Costaceae clade.Genera Monocostus K. Schum., Dimerocostus O. Kuntze, and Tapeinochilos Miq. were recovered as monophyletic while the genus Costus L. was found

File # 49ee
File # 49ee b DIVA optimizations shown with maximum areas per node limited to three (maxareas ϭ 3).AF ϭ Africa, SE ϭ Southeast Asia, ML ϭ Melanesia, AM ϭ South America.No nodes optimized as Central America (CA) although this area was included as a terminal distribution in the analysis.Bold ϭ date set as calibration point.''-'' indicates split of lineage between taxa on each side of em dash.
to be polyphyletic and is divided into three major clades: Costus, the Chamaecostus clade, and the Cheilocostus clade.The Chamaecostus and Cheilocostus clades correspond to lineages recovered with more complete taxon sampling of Costaceae (Specht in press) and are pending taxonomic recognition as genera (Specht and Stevenson in press).In that the topology obtained from this analysis reflected that of the larger and more completely sampled combined analysis, the chloroplast DNA phylogeny was used for the subsequent DIVA analysis and for estimating maximum likelihood branch lengths for the molecular clock analysis without need for topological constraints.

Estimating Branch Lengths and Assigning Rate Classes
A global likelihood ratio test indicated that the data fail to adhere to a global molecular clock (P ϭ 0.0001; Ϫ12711.193 with clock enforced, Ϫ12574.969without clock enforced).A local clock analysis was thus used to estimate ages of diversification in Costaceae.The F81uf ϩ I ϩ G model was determined to be the least parameter-rich model to fit the data (I ϭ 0.2726, G ϭ 0.8927).The topology from the maximum likelihood analysis is shown in Fig. 1, along with corresponding branch length estimates.Pairwise relative rates tests implemented in Hy-Phy indicate that the branches are optimally divided into four rate classes (0-3) as shown (Fig. 1).These rate classes are assigned to the branches as indicated for the local clocks analysis.

Biogeographic Analysis (DIVA)
DIVA analysis indicates potential ancestral distributions for key nodes in the Costaceae phylogeny (Table 2).The results shown in Table 2 and discussed throughout the remainder of this paper are those obtained when the maximum number of area optimizations for each node was constrained to three (maxareas ϭ 3), the condition which gave the most parsimonious solution (15 dispersal events) with the least number of optimal reconstructions for the basal-most node thereby minimizing the effect of wide distributions coded at the terminals.

Age Estimation
The resulting chronogram from the PAML local molecular clock analysis with the 105 mya calibration date is shown in Fig. 2. Exact node ages with standard error as determined by the local clocks algorithm in PAML are given in Table 2.With the exception of two pairs of terminal branches (the C. speciosus-C.globosus speciation event and Dimerocostus speciation), the use of 101 my and 109 my as calibration points did not increase the standard error beyond that around the 105 my age; therefore only the results of the analysis using 105 mya as the fixed calibration point are discussed further.
The first major cladogenic event discernable with extant taxa (Fig. 2 (Specht, in press).The ancestral distributions for all lineages within the Chamaecostus clade are South American, with all extant taxa having current ranges within South America.
Following the split-off of the South American lineage, the remaining Costaceae (Fig. 2: node F) existed in Southeast Asia and Africa and subsequently divided into two major lineages around 57 mya.Of these lineages, a DIVA-optimized vicariance event led to one lineage occupying Southeast Asia (Fig. 2: node G) and a second lineage colonizing Africa (Fig. 2: node I).The Southeast Asian lineage began to successfully diversify around 24 mya.At 11 mya (Fig. 2: node H) a dispersal event to Papua New Guinea (Melanesia) leads to a Melanesian ancestral distribution optimized for the monophyletic Tapeinochilos Miq.The Africa clade (Fig. 2: node I) begins to successfully diversify around 40.7 mya with a DIVA-optimized distribution in Africa only.This lineage remains in Africa and undergoes speciation in small numbers (see Fig. 2: node J) forming a basal paraphyly within the Costus clade until sometime around 33.7 mya, when a split (node K) occurs between an African clade (African melittophilous Costus; see Fig. 2: node L) and a New World clade (New World Costus radiation; Fig. 2: node M).The ancestor of this Africa-New World split has a distribution in Africa alone, requiring a dispersal event at this time to South America.All migration to Northern South America (e.g., Fig. 2: node N) and Central America occurs at the tips within the New World Costus radiation clade (no nodes are optimized with Central America as the ancestral distribution; see Table 2) and is considered to comprise multiple dispersal events and potential sympatric speciation events within various Central and South American populations.A more complete sampling of this clade is required to further investigate the details of neotropical migrations.

DISCUSSION
Recent uses of molecular clocks in combination with current phylogenetic hypotheses have enabled researchers to address temporal questions using genetic divergence (Sanderson 1997;Bremer 2000Bremer , 2002;;Yang andYoder 2000, 2003;Thorne and Kishino 2002;Yoder and Yang 2004).Phylogenetic dating using molecular sequence data requires at least one reference node to be fixed in order to calibrate the clock, converting the relative ages based on estimated branch lengths to absolute ages of lineage divergence and species diversification.The clock is often calibrated using fossil taxa or biogeographic events with known ages that can be assigned to a given node in the phylogeny.For Costaceae, no fossils have been identified.The date used for calibrating the molecular clock is thus inferred from the analysis of Zingiberales (Kress and Specht in press), using the divergence date between Zingiberaceae and Costaceae (105 mya) as the calibration date for converting relative to absolute rates.
The local clock method used in this analysis incorporates models that account for heterogeneity in different data partitions (Yang 2003;Yang and Yoder 2003), enabling the use of multiple gene partitions in a combined evidence analysis.When using single genes for determining rates or branch lengths, the confounding nature of rate and time are not accounted for, i.e., there is no way to distinguish between a high rate of sequence change over a short period of time and a low rate of divergence over a long period of time.Because the divergence times reflect the lineage divergences of unique organisms, they are shared for each gene region analyzed despite the inherent variation in evolutionary rates of the individual genes.Thus, the confounding problem of rate vs. time is potentially overcome by using multiple gene loci in a combined analysis that accounts for rate heterogeneity in the partitions, and the common parameter (i.e., divergence time) emerges from the combined data.

On Dating Costaceae
The use of 105 mya as the date for calibration of the Costaceae chronogram is based on two criteria.First, the dating of Zingiberales by Kress and Specht (2006) used rbcL, atpB, and 18S data, all of which are slowly evolving genes that do not fully resolve the Costaceae data set at the species level.The lack of sequence diversity between the four Costaceae exemplars used in Zingiberales can result in an underestimation of divergence times within the family.While the data set used for the Zingiberales analysis is appropriate for examining ages of diversifications among the major family lineages, within individual families it is more appropriate to use a molecular system that provides sufficient informative characters upon which to base a relative rates test and branch length estimation.
The use of the divergence time between Costaceae and Zingiberaceae (105 mya) as the calibration point allows the internal nodes of Costaceae to be dated based on branch lengths estimated the chloroplast DNA clock.Thus the ages of diversification within the family are estimated from the branch lengths of a more thorough sampling of Costaceae (37 rather than 4 taxa) and a molecular data set with sufficient character variation to resolve the appropriate specieslevel relationships within the family (as per Specht et al. 2001;Specht in press).Furthermore, the 105 mya date is closer topologically to the dates of calibration for the Zingiberales analysis (43 mya Ensete oregonensis S. Manchester & W. J. Kress fossil: Manchester and Kress 1993).It has been shown that closer proximity to the calibrated node yields greater fidelity of the age estimation results, therefore the 105 mya divergence date will be more accurate than the 52 mya diversification date (Yang et al. 1995

Costaceae Diversification in Space and Time
The ancestral distribution of Costaceae-Zingiberaceae is optimized by DIVA as Africa, Southeast Asia and South America.The first divergence within Costaceae (Fig. 2: node A) led to a division between a South American and an African ϩ Southeast Asian lineage, dating to 65 mya with a distribution identical to that of the ancestral population (AF, SE, AM).A second divergence separating an African and Southeast Asian ancestral population (Fig. 2: node F) occurred about 52 mya.The Asian clade appears to have diversified in SE Asia (Fig. 2: node G) starting around 23 mya.A dispersal to New Guinea approximately 11 mya (Fig. 2: node H) resulted in the Tapeinochilos lineage radiation.Meanwhile, the African lineage spread throughout the moist forest habitat of tropical Africa (Fig. 2: node I), forming several distinct clades each with only a few surviving taxa (Fig. 2: African Costus grade).About 34 mya however, a split occurs between an African lineage and a neotropical lineage (Fig. 2: node K).The African lineage continued to diversify in Africa (Fig. 2: node L), forming what is today a clade of bee-pollinated taxa with approximately twelve species (Fig. 2: African melittophilous Costus).Part of this lineage, potentially a single taxon, is likely to have dispersed to the neotropics from Africa.The resulting lineage (see Fig. 2: node M and ''New World Costus radiation'' clade) comprises the majority of neotropical species currently found throughout Central and South America.
Most of the species in the New World had evolved by the time of the Andean uplift (10 mya), thus major geographic changes in the region do not seem to be largely involved in speciation events within the neotropics.Within the New World Costus radiation, Central American and Caribbean species are sister to South American taxa (data not shown), indicating short distance dispersal to these areas from a core distribution along the base of the Andes and the Amazon and Orinoco basins.A more detailed investigation of this group, requiring rapidly evolving genetic markers to distinguish between species and populations, will yield more specific results about the patterns of speciation in this group and how they correlate with the historical geography of South and Central America and the Caribbean.

Are Costaceae a Gondwanan Family?
The current distribution of Costaceae is pantropical, a distribution often associated with Gondwana as the result of vicariance during the continental break-up.However, the dates of major early Costaceae cladogenesis (starting at 65 mya) do not suggest that the major lineages of Costaceae were formed coincidentally with the breakup of the supercontinent.This suggests that long distance inter-continental dispersal is more likely to have played a role in the formation of the major Costaceae lineages, with subsequent secondary dispersals along with localized speciation events leading to the levels of diversification we see today.
According to modern theories of geological activity and continental drift, the Gondwanan supercontinent began to separate around 180-150 mya (Hallam 1994;Scotese et al. 1988).At this time Africa and South America remained together forming a unified western portion of Gondwanaland, but were separated from the Eastern landmass that included Australia, Antarctica, Madagascar and India.Subsequent rotation of continental plates provided a land bridge connection between South America and Australia via Antarctica that remained until the Tertiary, approximately 65 mya (Hallam 1994).South America eventually became isolated from Australia at the end of the Eocene (55-35 mya) following the deterioration of connections across Antarctica.Africa and South America geologically split from one another around 110-90 mya (Sanmartı ´n and Ronquist 2004), however recent studies show that biotic exchange of flora and fauna was prolific until at least 80 mya due to the proximity of the continents and the shallowness and low salinity of the separating body of water (Lundberg 1993).The 105 mya separation between Zingiberaceae and Costaceae occurred during a time when exchange between the South American and African continents was likely.It is possible then that the early ancestors of Costaceae may have been present across the Western portion of Gondwana.The final break-up of Africa and South America could have led to the vicariance of lineages within Costaceae eventually leading to the separation and independent diversification of the South American and African lineages within the family, although the data with extant taxa alone indicate that divergence between these two lineages (Fig. 2: node A) does not initiate until 65 mya, long after the continents had separated.This could be an effect of unsampled (extinct) taxa that may have diverged earlier than 65 mya.
A potential scenario for the early cladogenesis of Costaceae (Fig. 2: node A) would be that the ancestral Costaceae lineage separated from the Zingiberaceae lineage in western Gondwana (AM and AF).A portion of the African lineage would then have given rise to the Asian population, potentially via the collision of India with Southeast Asia, provided that a population of Costaceae had dispersed to the continent of India as it rafted north and collided with Southeast Asia approximately 40 mya.A Laurasian migration via a North Atlantic land bridge during the late Eocene (Tiffney 1985), a route recently discussed for the migration of various lineages of tropical plants with modern pantropical distribution (Davis et al. 2002) and for Zingiberales (Kress and Specht 2006), could also explain the dispersal of an African clade to Southeast Asia.Fossil zingiberalean taxa present in North America during the Cretaceous and the Eocene (Friis 1988;Manchester and Kress 1993) lend credence to a scenario involving a northern migration.
During the late Eocene (up to 40 mya), periodic inundation of the Malaysian and Indonesian islands due to increased sea levels would have resulted in a lack of continuity in the Asian lineages and could have promoted isolation and extinction of local populations inhabiting small islands.This changing landscape makes it difficult to determine which direction Asian Costaceae migrated; however it does appear to explain the presence of highly localized taxa and morphologically variable species complexes (e.g., Costus globosus) as well as the relatively recent diversification date (22 mya;   2005).An early member of this bee-pollinated clade is likely to have dispersed to tropical America, initiating the New World Costus radiation (Fig. 2: node M). has been recorded for the flora of New Guinea during the Miocene as a result of the rapidly changing landscape of the island, and the quaternary has been important to subsequent speciation and diversification of lineages (Whitmore 1987).Upon its emergence, New Guinea quickly developed a diversity of tropical lowland and fragmented montane rainforest habitats as volcanic eruptions and subducting plate activity to the north quickly built a series of mountain ranges in the northern coastal area.Tapeinochilos, with its current center of diversity in Papua New Guinea, is likely to have evolved on this island.The increasing habitat diversity and resulting geologic barriers to gene flow may have played a role in the radiation of the Tapeinochilos lineage following its arrival in New Guinea.

New
The current distribution of Tapeinochilos throughout Papuasia (the biogeographic unit extending from New Guinea to the Solomon Islands and including the Bismarck Archipelago) is likely the result of inter-island dispersal from New Guinea to the surrounding archipelagos.The fruit of Tapeinochilos is an indehiscent capsule with a thin wall that becomes slightly swollen and fleshy when ripe (Gideon 1996).The fruit remains attached to the inflorescence stalk, supported by the subtending bract, until either the inflorescence disintegrates or the seeds are removed by animal dispersal agents (i.e., birds, small mammals) that are attracted to the seeds by the small but fleshy aril.The arillate seeds have a hard testa that is likely to survive ingestion, thus bird dispersal is a likely scenario for explaining the diversity of Tapeinochilos on the various island chains of Melanesia.Reports of bird dispersal for Costaceae include seeds of Costus allenii, C. laevis Ruiz & Pav. and C. guanaiensis Rusby in Panama (Schemske 1983).In addition, birds have been reported as effective seed dispersers of other Zingiberales taxa in New Guinea (Lamothe 1979;Beehler 1983) and Australia (Stocker and Irvine 1983;Barker and Vestjens 1989).While ant dispersal is also reported for Zingiberales (Horvitz and Beattie 1980;Horvitz and Schemske 1986), the indehiscent capsule of Tapeinochilos makes the seeds inaccessible to ants.Floating dispersal is another option, and could be possible based on the indehiscent capsule.The tolerance of Tapeinochilos fruits and seeds to periods of immersion in salt water has not been investigated.
Borneo has most recently been interpreted as an early Cretaceous island arc that was thrust on to the Asian margin in the mid-Cretaceous (McElhinny et al. 1974;Mitchell 1993;Metcalf 2002), and thus not rifted from Gondwana as a continental fragment as previously thought.
File # 49ee clear morphological and ecological diversity inherent in this group could have played a role in the evolution of the Tapeinochilos lineage.Costus sopuensis P. J. M. & H. Maas, known only from two collections from Celebes (Sulawesi, Palu) but presumed to be part of the Cheilocostus clade, may provide additional evidence of patterns of range expansion and diversification of the Southeast Asian lineages.Efforts to obtain sequence data from this taxon will help to determine its role in the biogeography of Asian Costaceae.

CONCLUSION
Modern pantropical patterns of species distribution within lineages tends to be interpreted as resulting from the historic fragmentation of the ancient landmass of Gondwanaland, regardless of whether or not the lineage predates the late Cretaceous breakup of the southern supercontinent.This analysis shows that while several lineages of Zingiberales, including the common ancestor of the Zingiberaceae-Costaceae lineages, were likely occupying the tropical areas of Gondwana prior to its breakup, the vicariance of Gondwana was not likely to have played a large role in the major diversification of modern Costaceae.While the earliest lineages may have been influenced by the fragmentation of Africa and South America, the current pantropical distribution of the family is better explained by a series of more recent local vicariance and dispersal events.It is possible that early Zingiberales had a more northern distribution, and that the current southern distribution is a result of the migration of tropical habitat from geographic areas that are now located in the Northern Hemisphere to its current equatorial position across northern South America, Africa and Southeast Asia.In this event, Costaceae as a lineage would have benefited from Asian island dynamics to spread to New Guinea and develop the Tapeinochilos lineage, and to spread into Africa.Dispersal to America across the spreading Atlantic Basin, first from a formerly tropical southern Europe (Chamaecostus clade) and second from tropical Africa (New World Costus radiation), would then explain the two dispersal events to the New World.These types of hypotheses can be further tested with additional information about biogeographic patterns that show similar patterns of migration over time.

Fig. 1 .
Fig. 1.-Phylogram of Costaceae based on analysis of chloroplast DNA data.Branch lengths estimated from combined maximum likelihood analysis of 37 trnL-F and 37 trnK (including matK) sequences under the F81uf model of evolution (Felsenstein 1981).Similarly colored branches indicate single rate classes as assigned for PAML local molecular clock analysis.Rate classes were based on the results of multiple pairwise relative rates tests (implemented in Hy-Phy [Kosakovsky-Pond and Muse 1998-2003]).

Fig. 2 .
Fig. 2.-Chronogram of Costaceae based on local clock analysis of chloroplast DNA data with calibration point set as the split between Costaceae and Zingiberaceae (105 mya) based on Kress and Specht (2006).Nodes are arranged according to the age estimates obtained using a local clock method for determining ages of diversification.Mya ϭ million years ago.Lettered nodes are referred to in text and Table 2. Node marked with * corresponds to the appearance of the derived hummingbird pollination morphology in New World Costus.Generic abbreviations are: C. ϭ Costus, M. ϭ Monocostus, D. ϭ Dimerocostus, T. ϭ Tapeinochilos.
Fig. 2: node G) for the Asian taxa relative to the diversification of the sister African clade (40 mya; Fig. 2: node I).The genus Costus.-TheCostus clade (including New World Costus radiation, the African melittophilous Costus, and the File # 49ee World Costaceae and pollination.-Duringthe mid-Tertiary (50-25 mya), the Costaceae flora of the New World consisted of a single South American clade that began to diversify around 49 mya (Fig.2: node B).This clade currently comprises approximately 12 species all maintaining a plesiomorphic generalist-pollinated floral form, including Monocostus (1 sp.), Dimerocostus (3 spp.), and the Chamaecostus clade (8 spp.).A dispersal event from Africa around 34 mya (Fig.2: node K) led to the establishment of New World Costus radiation lineage, which began to diversify approximately 22 mya (Fig.2: node M).The New World Costus radiation taxa all have flowers that are specialized for either bird or bee pollination(Specht 2005, in press).Historically, then, taxa from the African-dispersed clade quickly outnumbered the generalist-pollinated lineage already established in the neotropics: species radiation within the New World Costus clade began around 22 mya (21.71Ϯ 2 mya) with both bird and bee-pollinated floral forms being represented by 20 mya (Fig 2: *) while the basal South American clade remained relatively species-poor, based on the analysis of extant taxa.The current higher species diversity of the New World Costus radiation clade (ca.68 species) relative to that of the basal South American clade (12 species) may be due in part to the close association of the New World Costus radiation clade with specific pollinators, and the relative ecological success of this strategy for survival, dispersal, and diversification.Species-level diversification occurred rapidly in the New World Costus radiation clade.The Asian Costaceae: Tapeinochilos and Cheilocostus-clade lineages.-Thedate of the origin (19 mya) and diversification (11 mya; Fig. 2: node H) of the Tapeinochilos lineage corresponds well with the known geology and geography of the New Guinea region.The emergence of the island of New Guinea occurred around 25 mya in the late Oligocene-early Miocene, with the current tropical habitat developing around 15 mya (Whitmore 1987; Metcalf 2002).Major speciation Species currently found in Borneo are Costus paradoxus of the Paracostus clade (not included in this analysis), and C. globosus and C. speciosus of the Cheilocostus clade, all of which form part of the Asian Costaceae clade.Based on the geological and temporal context, population of Borneo by C. globosus and C. speciosus would have occurred via dispersal, likely from populations in India or southern Indochina where Costus speciosus is currently found in conjunction with C. globosus and C. lacerus Gagnep.The C. globosus complex extends from mainland Southeast Asia (China and Thailand) through the Malay Peninsula to Java and Borneo and across Sarawak to New Guinea.The diversity of this complex could have resulted from the movement (short-distance dispersal) of taxa across Malaysia from tropical India or Africa, and the

Table 1 .
Voucher and distribution information for species used in analysis.Living vouchers were sampled when herbarium vouchers were not available.Living collection localities are Harold L. Lyon Arboretum (L), New York Botanical Garden greenhouses (NY), Smithsonian greenhouses (NMNH), unless otherwise identified.

Table 1 .
Continued.Tapeinochilos tive ages to absolute dates of diversification.Although many fossil taxa have been found in Zingiberales lineage, no fossils have been assigned to Costaceae.The calibration point used in this analysis is a secondary calibration age, taken from the parallel local clocks analysis of Zingiberales ex O. G. Pet.) Maas

Table 2 .
Results from Local Clock and DIVA analyses for major nodes and subclades of the Costaceae phylogeny based on maximum likelihood analysis of plastid DNA sequence data.
a Node letter corresponds to Fig.2.
: node A) dates back to the late Cretaceous ca.65 mya.DIVA places the ancestral distribution in South America and Africa.The two clades resulting from the initial ;Yoder and Yang 2000; K. Bremer pers.comm.) African Costus grade: Fig.1, 2) is by far the largest lineage in the family with a total of approximately 85 species distributed in Africa and the neotropics.The stem age of the Costus lineage is around 56 million years old (Fig.2: node F) with the divergence of the African Costus lineage and the Asian lineage.Costus began to diversify around 40.