Genetic diversity and geographic structure in Aglaia elaeagnoidea ( Meliaceae , Sapindales ) , a morphologically complex tree species , near the two extremes of its distribution

Aglaia elaeagnoidea is the most widespread and one of the more morphologically diverse complex species in the largest genus of the mahogany family (Meliaceae, Sapindales). We performed maximum parsimony, maximum likelihood and Bayesian analyses (nuclear ITS rDNA) to estimate genetic relations among samples of Aglaia elaeagnoidea, and their phylogenetic position within Aglaia (more than 120 species in Indomalesia, Australasia, and the Pacific islands). Based on 90 accessions of Melioideae (ingroup) and four taxa of Cedreloideae (outgroup), this study 1) provides a first assessment of the genetic diversity of Aglaia elaeagnoidea; 2) investigates the geographic structure of the data in selected eastern and western regions of its distribution; and 3) suggests that Australia has been colonized only recently by A. elaeagnoidea and other species within the genus (Miocene/Pliocene boundary to Pliocene). Based on DNA data, morphology and additional evidence derived from biogenetic trends (secondary metabolites), the name Aglaia roxburghiana could be reinstated for specimens from the western end (India, Sri Lanka), but we have no data yet to indicate definitely where A. roxburghiana ends and A. elaeagnoidea begins either morphologically or geographically. Viewed in a more general context, Aglaieae are an ideal model group for obtaining more insights into the origin and evolution of Indomalesian and Australian biotas.

provided the first evaluation of these taxonomic concepts with data independent of morphology.Their study based on maximum parsimony and Bayesian analyses of nuclear ITS and plastid rps16 intron, as well as comparison of chemical profiles observed by means of high performance liquid chromatography (HPLC) and gas chromatography coupled to mass spectrometry (GC-MS), indicated that variable and complex species were more heterogeneous, i.e., probably containing more than one taxon each, than taxonomically isolated species (Muellner et al. 2005).
We performed maximum parsimony (MP), maximum likelihood (ML) and Bayesian analyses of sequence data from the internal transcribed spacers (ITS) of nuclear ribosomal DNA (nrDNA), defined as the unit containing the ITS1 spacer, 5.8S rRNA gene, and ITS2 spacer, to estimate genetic relationships within one of the morphologically complex and geographically most widely distributed species of the genus, Aglaia elaeagnoidea (A.Juss.)Benth.The aims of our study were to 1) provide a first assessment of the genetic diversity of A. elaeagnoidea by sampling near the two extremes of its distribution range; 2) investigate the geographic structure of the data over the sampled geographic range; and 3) infer the geographic and temporal origin of selected clades within Aglaia.The DNA data are compared to chemical profiles of the respective specimens to investigate if phylogenetic relationships are reflected by biogenetic trends.The ITS region was the DNA marker of choice since previous analyses of plastid rbcL, matK, and nuclear 26S rDNA did not provide sufficient phylogenetic information for Melioideae (compare Muellner et al. 2003Muellner et al. , 2006)), and screening with other plastid markers (e.g.rps16 intron, IGS atpB-rbcL, trnL intron and IGS trnL-trnF) did not exhibit a sufficient amount of informative sites.In addition, single copy nuclear genes, in contrast to ITS, are at present technically demanding to sequence and generally not retrievable from herbarium samples, because their amplification is highly subject to DNA quality (Cowan et al. 2006).This is especially true for tropical plants with high amounts of bioactive and/or PCR-inhibitory compounds, like the Meliaceae (A.N.Muellner pers.obs.).

MATERIAL And METHodS
We analyzed ITS data of 90 accessions of Melioideae, including representatives and type species of all three currently recognized sections within Aglaia (Amoora, Neoaglaia, and Aglaia), 13 accessions of A. elaeagnoidea from different locations of its distribution range (Fig. 1), and four accessions of Cedreloideae as outgroup taxa (Table 1).Broadscale intrafamilial phylogenetic relationships of Meliaceae were previously assessed using DNA sequence data from three regions: plastid genes rbcL, matK, and nuclear 26S rDNA (Muellner et al. 2003(Muellner et al. , 2006)).Phylogenetic relationships within Melioideae have recently been assessed using nuclear ribosomal DNA sequence data (Muellner et al. 2008b).

Plant material
Material of Aglaia elaeagnoidea was collected during excursions to Sri Lanka, Bangladesh, Thailand, Vietnam and Australia.Herbarium vouchers of this species are deposited at WU; herbarium specimens of the other Meliaceae are deposited at FHO, FR, FRIM, K, NCU, and WU (Table 1).ITS sequences of 35 taxa of Aglaia (other than A. elaeagnoidea) and 46 other taxa were available from the first authors' previous work on Meliaceae (Muellner et al. 2005(Muellner et al. , 2008b)).Besides 48 accessions of Aglaia, our ITS matrices also include representatives of all other genera of tribe Aglaieae (Lansium Corrêa, Reinwardtiodendron Koord., Aphanamixis Blume, Sphaerosacme Wall.ex Royle), plus representatives of all other tribes of Melioideae.
Folding predictions of secondary structures of the ITS1 and ITS2 RNA transcripts were made at the M. Zuker web server (http://www.bioinfo.rpi.edu/zukerm/) by use of the mfold program v3.1 (Mathews et al. 1999, Zuker et al. 1999).Foldings were conducted at 37 °C.After a first rough alignment with CLUSTAL version X (Thompson et al. 1997), corrections were made manually by using secondary structure predictions of ITS1 and ITS2 RNA transcripts as a guide for alignment across genera.Secondary structure predictions were confirmed by hemi-compensatory base changes and full compensatory base changes that preserved the predicted folding pattern.A total of 803 aligned positions were included in the matrices for phylogenetic analyses of ITS (including ITS1, 5.8S rDNA, and ITS2).
Gaps were coded as missing data.Aligned matrices are available from the first author (alexandra.muellner@senckenberg.de); new sequences have been deposited in GenBank under the accession numbers EU340981-EU340991 (http://www.ncbi.nlm.nih.gov/).

Phylogenetic analyses
MP analyses of the ITS dataset were performed using PAUP* v4.0b10 (Swofford 2002).Substitutions at each nucleotide position were treated as independent, unordered, multi-state characters of equal weight (Fitch parsimony, Fitch 1971).Heuristic searches were performed using 1 000 random additions of taxa, tree bisection-reconnection (TBR) branch swapping, and the option MulTrees (keeping multiple, shortest trees), but holding only 10 trees per replicate to reduce time spent in swapping on large numbers of trees.After 1 000 replicates, we then used the shortest trees found as starting trees for a swapping-to-completion search (but with a tree limit of 15 000).Robustness of clades was estimated using bootstrapping (Felsenstein 1985) with 1 000 replicates, using simple sequence addition, TBR branch swapping, and MulTrees, again holding 10 trees per replicate.controls the magnitude of autocorrelation per million years (Myr) along the descending branches of the tree.Prior gamma distributions on parameters of the relaxed clock model were as follows (following the 'multidivtime' manual for setting the mean and SD; http://statgen.ncsu.edu/thorne/multidivtime.html): the mean and SD of the prior distribution for the root age were set to 75 Myr based on fossils (below) and previous studies of Muellner et al. (2006Muellner et al. ( , 2007Muellner et al. ( , 2008a)).The mean and SD of the prior distribution for the ingroup root rate were set to 0.0024 substitutions/site/Myr by dividing the median of the distances between the ingroup root and the tips by 75 Myr.The prior and SD for ν were set to 0.013, following the recommendation of Thorne's manual that the time between root and tips multiplied by ν be about 1. Markov chains in 'multidivtime' were run for 1 million generations, sampling every 100th generation for a total of 10 000 trees, with a burn-in of 10 000 generations before the first sampling of the Markov chain.

Constraints and calibrations
Absolute time estimates in the Bayesian approach were obtained by simultaneously constraining four nodes (numbered 1-4 below); for the NPRS clock we constrained one node (numbered 5 below).

The clade comprising
For absolute ages we relied on the geologic time scale of Gradstein et al. (2004).

RESuLTS
We checked the ITS sequences for the presence of structural motifs and secondary structure.This enabled us to unambiguously align the entire ITS region across all genera (ingroup and outgroup), without losing any potentially informative characters for subsequent phylogenetic analyses.There was no indication for multiple copies of ITS in any individual included in our analysis, indicating no need to clone PCR products.Sequences of all ingroup and outgroup taxa contained the conserved angiosperm motifs in ITS1, 5.8S and ITS2 (Liu & Schardl 1994, Hershkovitz & Zimmer 1996, Jobes & Thien 1997).

Phylogeny estimation
Alignment of all ITS region sequence positions resulted in a matrix of 803 characters for the whole set of taxa.The length of the total matrix including 18S and 26S rDNA flanking regions was 1 010 characters.For the entire ITS region, 517 (64 %) positions were variable and 418 (52 %) were potentially parsimony informative.The parsimony search produced 288 most parsimonious trees of 2 863 steps with a consistency index (CI) = 0.34 and a retention index (RI) = 0.61.Fig. 2 shows the strict consensus obtained from the MP analysis of 90 ingroup and four outgroup taxa, and bootstrap values.The Bayesian (Fig. 3) and ML (Fig. 4) analyses were all based on the same dataset as used for the MP analysis.The Bayesian tree was based on a total of 20 000 trees and a burn-in of 910 trees (Fig. 3).

Infra-specific polymorphism
We included 13 accessions of the morphologically complex species Aglaia elaeagnoidea in our analyses, from two main areas beyond the western and eastern boundaries of Malesia.The samples exhibited high genetic diversity, topologically covering two portions of our phylogenetic trees (Fig. 3, 4).Accessions from western regions showed the following pattern: samples from Thailand and Vietnam formed a strongly supported clade in all trees (98 % bootstrap percentage, BP, Fig. 2; 100 % posterior probability, PP, Fig. 3; 97 % BP, Fig. 4), followed by a grade of samples from Bangladesh and Sri Lanka (Fig. 2-4); all western samples together formed a weakly to moderately supported clade (73 % BP, Fig. 2; 93 % PP, Fig. 3; 61 % BP, Fig. 4; in the following referred to as 'Western Group').Accessions from eastern regions appeared in a strongly supported clade (100 % BP and PP, respectively, in all trees; in the following referred to as 'Eastern Group').

Divergence time estimation
Divergence time estimates for selected nodes are given in Fig. 3. Results based on NPRS and the Bayesian approach yielded similar age estimates.Age estimates derived from the Bayesian approach, based on several fossils, can be viewed as best approximations.Our Bayesian age estimates suggest that the clade of Aglaia samoensis and A. sapindina dates back to 5.7 ± 3.7 Ma, the clade uniting samples of the Western Group to 11.7 ± 4.8 Ma, the clade uniting samples of the Eastern Group to 5.3 ± 3.5 Ma, and the clade of A. australiensis and A. meridionalis to 5.2 ± 3.4 Ma (Fig. 3).

dISCuSSIon
Our study 1) provides the first assessment of the genetic diversity of Aglaia elaeagnoidea on the basis of a geographic sampling from the two ends of the distribution; 2) investigates the geographic structure of the data over the sampled geographic range; 3) evaluates the two ends of the morphological continuum which led to the wide species concept formerly adopted for A. elaeagnoidea; and 4) suggests that Australia has been colonized only recently by A. elaeagnoidea and other species.
Our conclusions are based on several lines of evidence.

Genetic diversity and geographic structure of Aglaia elaeagnoidea
Aglaia elaeagnoidea is a widespread species, which occurs throughout the range of the genus except for the Solomon Islands and extends beyond the range of any other species in the genus in Western Australia.It appears to have two distinct entities at the eastern and western ends of its range, outside the Malesian area, linked by a series of morphological intermediates through Malesia.It was therefore considered to be a complex species by Pannell (1992).Unlike most species of Aglaia, it is not a rain forest tree, apparently adapted to a more seasonal climate.It is usually coastal, especially in the east of the range, but is also found inland and at higher elevations in India, Sri Lanka, Thailand and Borneo, on dry sandy soils and limestone.
On the basis of our geographic sampling strategy, specimens assigned to A. elaeagnoidea based on morphological grounds according to the current literature (Pannell 1992(Pannell , 2004)), appear as genetically heterogeneous (Fig. 2-4).Specimens to the far east and west of the range of the species belong to two different genetic entities based on ITS data.

Taxonomic considerations -Aglaia elaeagnoidea and A. roxburghiana
Historically, two main species were recognized in the Aglaia elaeagnoidea complex: A. elaeagnoidea (basionym Nemedra elaeagnoidea A.Juss.1830) from Australia and A. roxburghiana (Wight & Arn.) Miq.(basionym Milnea roxburghiana Wight & Arn.1834) from India and Sri Lanka.Various varieties and species names were applied to both of these to deal with the variation encountered within and between these two extremes in morphology and geographical range.In the monograph of Aglaia (Pannell 1992), it was treated as one widespread species, A. elaeagnoidea.It is usually a small tree or shrub, 5-10 m tall.The large pale orange or almost white peltate scales are characteristic, but they vary in colour, in the presences and extent of the fimbriate margin and in their density on the plant.In the eastern part of the range, the scales are larger, paler in colour and more frequently entire than in the west, the density is greater on the leaves, but the fruits may have fewer scales than in the west and the aril may be vestigial.In India, A. elaeagnoidea is sometimes a larger tree, to 20 m, than in the rest of the range of the species.Large-leaved specimens from the west can also be almost indistinguishable from another complex species, A. edulis, unless fruits are present; these are much larger (up to 3.2 × 3.8 cm) in A. edulis than are those of A. elaeagnoidea (up to 2 × 1.5 cm) and usually 3-locular rather than 2-locular.The molecular data presented here confirm that the two extremes of variation in A. elaeagnoidea can be distinguished at the molecular level, just as they can on morphological characters.Interestingly, samples from Australia are not in the clade which separates most eastern species of section Aglaia from the rest of the genus (top clade in Fig. 2-4).The material of A. elaeagnoidea from Australia was collected from a relatively small area of Queensland and the samples are all similar at the molecular level for the DNA region sequenced.There is more variation in the samples from the west of the range, reflecting the larger area from which they were collected.Specimens from these two widely separated regions are distinct at the morphological and at the molecular level.On morphological and molecular evidence, the name Aglaia roxburghiana could be reinstated for specimens from the western end, but we have no data yet to indicate clearly where A. roxburghiana ends and A. elaeagnoidea begins either morphologically or geographically.This will be subject to further investigations.To reach a detailed appreciation of A. elaeag-noidea in addition to the patterns already elucidated by present evidence, sampling of additional specimens, especially from the Malesian region, will be necessary.

Additional evidence derived from biogenetic trendspatterns of secondary metabolites in the Western versus Eastern Group of Aglaia elaeagnoidea
Broad-based phytochemical comparisons with the 13 samples of Aglaia elaeagnoidea collected in various parts of Thailand, Vietnam, Bangladesh, Sri Lanka, and Australia showed chemical differences clearly separating all collections from Australia (Eastern Group) from the others (Western Group).Especially the genus-specific flavaglines of all representatives of the Eastern Group deviated by derivatives with a dimethoxylated aromatic ring A, like rocaglamide and aglafoline, whereas all other collections from the west were characterized by a methylenedioxy and a methoxy group in ring A, like pannellin and aglaroxin A (Brem 2002, Hofer 2002).Additionally, the Australian samples were characterized by the formation of a triterpene derived limonoid, 6-α-acetoxygedunin (Hofer 2002).Although widespread in the Meliaceae, limonoids are apparently rare in Aglaia, only described so far for another collection of A. elaeagnoidea from Sempu Island (Java) also containing the flavagline aglafoline with a dimethoxylated ring A (Fuzzati et al. 1996).
The accumulation of different bisamides represents another important chemical character distinguishing the Australian samples from the Western Group.Whereas odorine derivatives are dominating in Australian collections, samples in the Western Group are characterized by piriferine derivatives.Both types of bisamides were also found as characteristic building blocks in the flavaglines (Greger et al. 2008).Different patterns of lignans and flavonoids may also contribute to a taxonomic segregation of the Eastern Group of A. elaeagnoidea (Hofer 2002, Greger unpubl.).

Temporal origin of Western and Eastern Groups of Aglaia elaeagnoidea, and relatives
Our results imply that the ancestor of the clade comprising the individuals of the Western Group currently assigned to Aglaia elaeagnoidea originated in the western part of the distribution range of the genus and of A. elaeagnoidea, respectively, subsequently dispersing eastwards.This pattern from west to east can be observed based on the topologies of the trees within the Western Group (Fig. 3, 4).These topologies suggest that earlybranching within A. elaeagnoidea occurred in western areas including Sri Lanka and Bangladesh.Aglaia elaeagnoidea then seems to have extended its range eastwards to areas including Thailand and Vietnam.
Based on the evidence of the tree topologies suggesting that the Western Group of A. elaeagnoidea originated in the very western part of the distribution range of the genus, we were interested in the relative timing of divergence events that had given rise to the western versus eastern clades of Aglaia species in our trees.If Aglaia had originated in the west, then clades comprising samples from Australia, as representatives of lineages from the east, should be comparably younger.Our divergence time estimates yielded estimates confirming these expectations (Fig. 3).Based on this measure, clades/groups with specimens collected in Australia (from species either restricted in their occurrence to Australia, like A. australiensis and A. meridionalis, or having a broader range of distribution, like A. sapindina; compare Fig. 3) date back to 5.2 -5.7 Ma.The Eastern Group of A. elaeagnoidea dates back to minimally 5.3 mya.In contrast, the group comprising western A. elaeagnoidea dates back to minimally 11.7 mya.This suggests that old Gondwanan land masses like Australia might have been colonized only recently by A. elaeagnoidea and other species (compare Muellner et al. 2008a), with divergence events mostly dating back to the Miocene/Pliocene boundary and subsequent Pliocene times.

Fig. 3
Fig.3 Fig.4 Guarea and Ruagea was constrained to minimally 23.03 Ma (the upper bound of the Late Oligocene), based on fossil pollen of Guarea from the Oligocene San Sebastian Formation in northern Puerto Rico (Graham & Jarzen 1969).3. The crown group of Melieae was constrained to minimally 20.43 Ma (the upper bound of the Aquitanian in the Early Miocene), based on fossil pollen of Melia (pollen similar to Melia azedarach) from the Early Miocene of Cameroon (Salard-Cheboldaeff 1978).4. The clade comprising Chisocheton and the remainder of Guareae and Aglaieae was constrained to minimally 5.3 Ma (the upper bound of the Late Miocene), based on Miocene fossil wood of Chisocheton (Chisochetonoxylon) from the Birbhum District in West Bengal (Ghosh & Roy 1979).

Table 1
Voucher information and GenBank accession numbers for samples used in this study (in alphabetical order).Voucher specimens are deposited in the following herbaria: FHO = Daubeny Herbarium, University of Oxford; FR = Herbarium Senckenbergianum, Senckenberg Research Institute; K = Royal Botanic Gardens Kew; KEP FRI = Forestry Institute Malaysia; KYGH = Kyoto Takeda Herbal Garden; NCU = University of North Carolina; WU = University of Vienna.
For the Bayesian analyses, model parameters were estimated directly during two parallel runs, using four simultaneous chains and two million cycles each, sampling one tree every 100 generations.Trees that preceded stabilisation of the likelihood value were excluded, and the remaining trees were used to calculate posterior probabilities via the construction of a majority rule consensus tree in PAUP.For the ML searches with RAxML we employed the GTR + G model, using 25 rate categories (instead of four as used in the Bayesian analyses).Locations and number designation of specimens sampled, and total distribution of Aglaia elaeagnoidea (small map on the right: after Pannell 1992).Outer line delimits the geographical distribution of the tribe Aglaieae.
(Thorne & Kishino 2002) 2003ap values moderate support and 85 -100 % strong support.ML analyses were performed with RAxML v2.2.1(Stamatakis 2006, http://icwww.epfl.ch/~stamatak/index-Dateien/Page443.htm), and Bayesian analyses with MrBayes v3.1.2(Ronquist&Huelsenbeck2003,http://mrbayes.csit.fsu.edu/).The substitution models employed in these analyses were found using Modeltest v3.06(Posada & Crandall 1998, http://darwin.uvigo.es/software/modeltest.html),which indicated the general time reversible model as best fitting our data with a proportion of invariable sites and a gamma shape parameter alpha to model rate heterogeneity (GTR + I + G). et al. 1998) was then used to calculate branch lengths and their variance, given the sequence data (94 ITS sequences of a length of 803 nt), the model parameter values from PAML, and the specified rooted topology.Branch lengths from 'estbranches' became the priors for mcmc searches in 'multidivtime'(Thorne & Kishino 2002)that sought to find the most likely model of rate change (with rate change assumed to be log-normally distributed), given the topology, time constraints on nodes (below), and a Brownian motion parameter (ν that Fig. 2 Strict consensus tree of 288 trees obtained from the maximum parsimony analysis of the ITS dataset of 94 Meliaceae accessions.Numbers are bootstrap percentages (1 000 replicates).