Phylogeny of Acronychia (Rutaceae) and First Insights into Its Historical Biogeography and the Evolution of Fruit Characters

Background The genus Acronychia (Citrus family, Rutaceae) contains 49 species of trees and shrubs that are found mainly in rain forest. The genus has a large distributional range from mainland southern Asia to Australia and New Caledonia, but most species are endemic to either New Guinea or Australia. This study aimed to provide the first detailed molecular phylogeny of Acronychia and use it to test the taxonomic value of fruit morphological characters, and infer the historical biogeography of the genus. Methodology Phylogenetic analyses (Bayesian Inference, Maximum Likelihood) were undertaken on nucleotide sequence data from two plastid (psbA-trnH, trnL-trnF) and three nuclear markers (ETS, ITS, NIAi3) from 29 Acronychia species (59% of the genus) and representatives of related genera. Results and Conclusions The results indicate that the South-East Asian genus Maclurodendron is nested phylogenetically within Acronychia and must be synonymized to render Acronychia monophyletic. Fruit morphological characters have been used previously to infer relationships within Acronychia and our analyses show that these characters are informative for some subclades but are homoplasious for the group as a whole. Apocarpous fruits are the ancestral state in Acronychia and subapocarpous and fully syncarpous fruits are derived. The unisexual flowers of Maclurodendron are derived from bisexual flowers, which are found in all species of Acronychia as well as its relatives. Acronychia probably first evolved on Australia with range expansion to New Guinea via stepping-stone dispersal or direct land connections within the Sahul Shelf, followed by two independent dispersals to areas west of New Guinea. Most species of Acronychia occur in either Australia or New Guinea, but no species occurs in both regions. This is surprising given the close proximity of the landmasses, but might be explained by ecological factors.

species of Acronychia occur in either Australia or New Guinea, but no species occurs in both regions. This is surprising given the close proximity of the landmasses, but might be explained by ecological factors.
the so-called Euodia-alliance, are possible close relatives of Acronychia [31]. Most of these genera have been sampled in recent molecular analyses [1,2], which showed that the Euodia-alliance is not monophyletic.
The genus Acronychia itself is probably not monophyletic-Maclurodendron is nested within it [1]. Maclurodendron was erected by Hartley [32] for three South-East Asian species he earlier ( [3] p. 469) excluded from Acronychia. The genus now consists of six species, which are distributed on the South-East Asian mainland, and from Sumatra and the Malay Peninsula east to the Philippines [32]. Maclurodendron is morphologically very similar to Acronychia and the only consistent differences between them are dioecious (Maclurodendron) versus hermaphroditic (Acronychia) flowers and imbricate (Maclurodendron) versus valvate (Acronychia) petals [32].
In this study we present the first detailed phylogenetic reconstruction of the genus Acronychia based on nuclear and plastid sequences from 59% of the species selected to cover the geographical range and morphological diversity of the genus. The major goals of this study were to (1) reconstruct the phylogenetic history of Acronychia, (2) identify the closest relatives of Maclurodendron within Acronychia in order to gain insight into the evolution of dioecy in the group and the geographic origin of Maclurodendron, (3) determine whether or not A. octandra is part of Acronychia despite its unusual fruit morphology, (4) evaluate the relevance of Hartley´s [3] evolutionary lines that are mainly based on the presence or absence of septicidal fissures, and the structure of the exo-and mesocarp for classification within Acronychia, and to (5) identify biogeographic patterns, and infer directions of dispersal and the geographical origin of the genus.

Taxon Sampling
In total, 29 of the 49 currently accepted species of Acronychia, which together comprehensively sample the morphological diversity and geographical range of the genus, were included in this study. These 29 species include 15 of the 19 Australian endemics (79%) and 10 of the 27 New Guinean endemics (37%). The sparser sampling of the latter is due to most species (52%) being known only from the type collection or from very few collections from the type locality, which is, in many cases, quite remote [3,5,9]. All widespread species (viz. extending beyond Australia and New Guinea) were sampled. Acronychia laevis, which occurs in eastern Australia and New Caledonia, was sampled from Australian material. One sample of the genus Maclurodendron, which was shown to belong to Acronychia [1], was included in the analysis. Euodia has been shown to be part of the clade sister to the main Melicope/Acronychia clade of Appelhans et al. [1] and so two of the seven species of Euodia (E. hortensis and E. montana) were used as outgroups. Several species of Melicope, Medicosma, Tetractomia and Comptonella, the closest relatives of Acronychia [1], were included to assess the monophyly of the latter.
Voucher information for all specimens used in this study is given in Tables 1 and 2.

DNA extraction, amplification and sequencing
Laboratory work was carried out in Cairns, Göttingen and Sydney. Leaf samples were taken from silica-dried specimens or from herbarium sheets. The plant material was ground in a Tis-sueLyser II (QIAGEN, Valencia, California, USA) using stainless steel beads. Total DNA was extracted using the Qiagen DNeasy Plant Mini Kit (QIAGEN, Valencia, California, USA) according to the manufacturer´s instructions (Cairns, Sydney). In the Göttingen laboratory, the original protocol was modified in two ways: the extraction was done without RNase and the lysis step was increased from 10 to 30 minutes to increase the efficiency of cell wall lysis. Five markers were amplified and sequenced for this study as these markers proved to be highly variable and informative in the closely related genus Melicope [1,33]. The three markers chosen from the nuclear genome were ITS (internal transcribed spacer), ETS (external transcribed spacer) and NIAi3 (the third intron from the single copy gene NIA; nitrate reductase) and the spacer regions trnL-trnF and psbA-trnH were chosen from the plastome.
PCR was performed in a 25μL reaction volume containing 2.5 μL of 10x PCR buffer, 2.5 mM MgCl 2 , 0.2 mM of each dNTP, 0.2 μM of each primer, 3 U of BioTaq (Bioline, London, UK) and 1 μL of template DNA. The amplification program used for ETS, ITS, psbA-trnH and trnL-trnF consisted of 5 minutes of initial denaturation at 95°C; 36 cycles of 1 minute denaturation at 95°C, 40 seconds of annealing at 53-54°C, an elongation of 40 seconds to 1 minute (depending on the length of the target sequence) at 72°C; and a final elongation for 5 minutes. A touchdown PCR program was used for the amplification of NIAi3 [34].
Amplified PCR products were cleaned using either the MSB Spin PCRapace Kit (Göttingen laboratory; Stratec Biomedical AG, Birkenfeld, Germany) or the SureClean Plus Kit (Cairns and Sydney laboratories; BioLine, London, UK) and then sequenced using the BigDye version 3.1 kit (Applied Biosystems, Foster City, California, USA) following the manufacturer´s instructions. The sequencing products were purified using ethanol precipitation and run on ABI 3100 (Göttingen) or ABI 3730 (Cairns, Sydney) sequencers.
Genbank accession numbers for all sequences used in this study are shown in Tables 1 and 2.

Multiple sequence alignment and phylogenetic reconstruction
Sequences were checked and edited in Geneious version 5.6.4 (Biomatters Ltd, Auckland, New Zealand) and then aligned in SATé [44] using the settings described in Appelhans et al. [33]. The alignments were checked and manually edited in Geneious version 5.6.4 and MacClade 4.08 (Sinauer Associates Inc., Sunderland, MA, USA), and are available from TreeBASE (study accession URL: http://purl.org/phylo/treebase/phylows/study/TB2:S17897). jModeltest v2.1.3 was used to determine the best-fitting model of sequence evolution for each single marker alignment under the Bayesian Information Criterion (BIC; [45,46]). The best-fitting models are shown in Table 3. Phylogenetic tree estimations were performed using Bayesian Inference (MrBayes 3.2.1; [47,48]) and Maximum Likelihood (ML; Garli 2.0; [49]). All single marker alignments were first analyzed separately and, since the trees showed no supported incongruences, the alignments of the five markers were concatenated and analyzed together. In order to evaluate the influence of the hybrid species A. littoralis [50] on the tree topology, we performed an additional Bayesian analysis excluding the A. littoralis specimens. Only specimens for which sequences of at least three out of the five markers could be obtained were included in these concatenated analyses in order to minimize the amount of missing data. Overall, sequencing of ETS and ITS had a higher success-rate and several species that were not included in the concatenated analyses were included in the ETS and/or ITS analyses. We therefore show and discuss the results of the ETS and ITS analyses in addition to those of the concatenated analyses. The Bayesian analysis of the single marker datasets and the concatenated dataset consisted of two independent Markov Chain Monte Carlo (MCMC) runs with four chains each. The length of the runs was set to ten million generations and a tree was saved every 100th generation. In order to evaluate if the two runs reached stationarity, the average standard deviation of split frequencies between the two runs was checked; a value of <0.01 was regarded as sufficient. The effective sample size (ESS) and the burn-in were determined using Tracer v1.6.0 [51]. Based on the results the first 10-15% of the trees were discarded as burn-in and the remaining trees were used to calculate a 50% majority-rule consensus tree in MrBayes 3.2.1.
The ML analyses consisted of five independent searches each comprising 1000 bootstrap replicates to obtain statistical support (bootstrap values, BS) for the tree topology. The models of sequence evolution used for the ML analyses were estimated as described above and are shown in Table 3. SumTrees v3.3.1, as implemented in the python package DendroPy 3.12.0 [52], was used to construct the bootstrap consensus tree.
The consensus trees from the Bayesian and ML analyses were edited in FigTree v1.4.0 (http://tree.bio.ed.ac.uk/software/figtree/) and rooted using the two Euodia species as outgroups. Clades with Bayesian posterior probability (PP) values of 0.95 were regarded as statistically supported. In the ML analyses, clades with a bootstrap (BS) value of 90% or higher were regarded as strongly supported, clades with a BS support of 70%-89% were regarded as having medium support, and clades with a BS support of 50%-69% were considered to have low support. Clades with less than 0.95 PP and 50% BS were regarded as not supported and treated as polytomies.

Tracing morphological characters and geographical areas
Two fruit characters-fusion of carpels, and texture of mesocarp-have previously been used to define 'evolutionary lines' within Acronychia [3]. The reliability of these characters as phylogenetic markers was critically assessed by plotting them on the Bayesian consensus tree of the concatenated matrix using parsimony in Mesquite v.3.02 [53]. Fruit character states were coded as follows: a) degree of fusion of carpels: 0-fruit completely syncarpous; 1-septicidal fissures extending from ¼ to ½ the length of the fruit; 2-septicidal fissures extending to at least ½ of the length of the fruit; 3-fruit nearly completely apocarpous (and carpels beaked); 4-other fruit types (dehiscent fruits in most taxa and drupaceous fruits with varying degrees of carpel fusion in Comptonella); b) texture of the mesocarp: 0-mesocarp drying semifleshy and/or not differing from exocarp; 1-mesocarp drying spongy-crustaceous; 2-mesocarp drying (sub) woody; 3-no evident mesocarp; 4-other fruit types. Most outgroup taxa have dry and dehiscent fruits, so that the texture of the mesocarp is not comparable to that of drupaceous fruits, and the fruits of Comptonella and Melicope can be fully syncarpous or almost apocarpous. In these cases the characters state 'other fruit types' was used.
To reconstruct ancestral areas and the geographical origin of Acronychia, geographical areas were plotted on the Bayesian consensus tree of the concatenated matrix using Mesquite with the settings described above. More sophisticated ancestral area reconstruction methods implemented in programs such as Lagrange [54,55] or BioGeoBEARS [56] require fully bifurcating and dated phylogenies. Since no fossils have been described for Acronychia or near relatives, such analyses are only feasible for a broader taxon sampling (Appelhans et al., in preparation) and are thus beyond the scope of this study. Geographic areas were coded as follows: 0-Eastern Australia; 1-New Guinea; 2-widespread. Several outgroup taxa are narrow endemics, e.g. Comptonella microcarpa (New Caledonia) and Melicope clusiifolia (Hawaiian Islands). However, both taxa were coded as having a widespread distribution since they are the sole representatives in this analysis of clades with wide distributions in the South-East Asian and Pacific areas [1].

Phylogeny and Systematics
The consensus trees from the Bayesian and the ML analyses of the concatenated five-marker dataset (Fig 1) are generally well resolved, well supported, and showed no supported incongruences. The backbone of the trees is highly supported with most clades showing high support of 0.95 PP and 90% BS respectively. Consensus trees from the single-marker analyses are generally less well resolved (ETS, Fig 2; ITS, Fig 3; trees for other markers not shown), but showed no supported incongruences (incongruent clades with 0.95 PP and 70% BS) with the consensus trees of the concatenated dataset.
Our results confirm that the genus Maclurodendron is nested within Acronychia, indicating that Maclurodendron needs to be synonymized under Acronychia for Acronychia to be monophyletic. The Acronychia clade (incl. Maclurodendron) is sister to a clade composed of Melicope clusiifolia, M. durifolia, and M. elleryana, which represent the two largest sections of Melicope: Lepta and Pelea. The Acronychia-Pelea-Lepta clade in turn is sister to a clade consisting of M. ternata, the type of Melicope, and the New Caledonian endemic genus Comptonella.
All these main clades are highly supported (0.99-1.00 PP and 90-100% BS) which supports the finding of Appelhans et al. [1] that Melicope is paraphyletic with respect to Acronychia, Comptonella and Maclurodendron.
Within Acronychia, five distinct clades are observed. Clade 1, consisting solely of the Australian species A. octandra, is sister to the remainder of Acronychia (0.71 PP, not supported in the ML analysis). Clade 2 (1.00 PP, 73% BS) contains samples of A. littoralis, A. oblongifolia and A. wilcoxiana, species which are also exclusively Australian. Clade 3 comprises A. acuminata, a rare, narrowly endemic species of North-East Queensland (Australia). Its position as sister to Clades 4 and 5 is resolved only by the Bayesian analysis, but without support (0.85 PP). Clade 4 is entirely Australian except for A. laevis, which is also present on New Caledonia. Clade 4 is resolved by both Bayesian and ML analyses, but is supported only in the ML analysis (0.94 PP, 75% BS). This large clade is further divided into three subclades. Clade 5 (1.00 PP, 99% BS) lacks Australian taxa and consists of all sampled New Guinean species, the two widespread species A. pedunculata and A. trifoliolata, and the genus Maclurodendron. Within this clade, A. pedunculata and Maclurodendron form a well-supported subclade (1.00 PP, 92% BS) that is sister to the rest of the clade. The second widely distributed species, A. trifoliolata, is most closely related to A. ledermannii and A. pullei (0.90 PP, not supported in the ML analysis), two rather widely distributed New Guinean endemics.
For several species more than one sample was included. Three of these species were found to be non-monophyletic with high support in both Bayesian and ML analyses. The three specimens of A. ledermannii form a clade with A. pullei. Acronychia suberosa was resolved as paraphyletic with A. pubescens. Acronychia littoralis was shown to be polyphyletic with two accessions in Clade 4 (related to A. vestita, A. imperforata and A. acronychioides) and one accession in Clade 2 (related to A. oblongifolia and A. wilcoxiana). In addition, A. pedunculata was resolved as paraphyletic with respect to Maclurodendron, although this topology was not supported in the Bayesian analysis (0.88 PP) and only weakly supported in the ML analysis (54% BS).
The additional Bayesian analysis of the concatenated data matrix excluding the samples of the hybrid species A. littoralis revealed one difference in the tree topology. The two accessions of A. oblongifolia were resolved as a clade with 0.95 PP, which was sister to A. wilcoxiana with 1.00 PP (results not shown), while one of the accessions of A. oblongifolia was sister to A. For several species only ETS and/or ITS sequences were obtained and they were not included in the combined analyses in order to minimize the effect of missing data on the results. The assignment of these species to clades within Acronychia was done using separate ETS and ITS analyses (Figs 2 and 3). In the ETS analysis (Fig 2) the two New Guinean Phylogeny of Acronychia (Rutaceae) species-A. foveata and A. cartilaginea-group with New Guinean species (Clade 5 in Fig 1). Acronychia foveata is most closely related to the widespread A. trifoliolata. Two other New Guinean species-A. emarginata and A. rugosa-group with the large Australian clade (Clade 4 in Fig 1), however, this placement was only supported in the ML consensus tree of ETS. Two Genbank accessions from Australia representing the species A. acronychioides and A. crassipetala were included in the ITS dataset. While A. crassipetala is part of Clade 4, the Genbank accession of A. acronychioides clusters with A. pedunculata, which is the most widespread species of Acronychia, but does not occur in Australia. The identity of the voucher (Elick 283, CNS) for the A. acronychioides ITS sequence has been confirmed by two of us (DC, MSA), but this surprising placement should be tested with data obtained in a different laboratory from additional, independent samples.

Fruit Evolution and Hartley´s Evolutionary Lines
Character mapping under parsimony indicates that fruit morphology is demonstrably homoplasious, except for the character states that are autapomorphic for A. octandra (Fig 4). Concerning the fusion of carpels, the large Australian clade (cf. Clade 4 in Fig 1) contained the whole array of character states found in the genus, excluding the autapomorphies of A. octandra. Despite the homoplasy, the degree of carpel fusion is an informative character in some subclades: the three early-branching clades within Acronychia (Clades 1 to 3; Fig 1) contain taxa with septicidal fissures only, whereas Clade 5 (Fig 1) comprises taxa with fully syncarpous fruits only. The character states for 'mesocarp texture' were scattered throughout the tree with no apparent pattern and all three states occur in the clades that contain more than one species (Clades 2, 4, 5; Fig 1).

Biogeographical insights
The Australian and New Guinean species are strictly separated in the phylogenetic analyses of the concatenated matrix ( Fig 5) and none of the five clades contain species from both regions. All early-branching clades are endemic to Australia and this area was reconstructed as the ancestral area of Acronychia (Fig 5). The New Guinean clade contains two widespread subclades (A. trifoliolata; A. pedunculata + Maclurodendron) which are not closely related, therefore two independent colonization events to western Malesia and South-East Asia are inferred (Fig 5).
The separate ETS and ITS analyses shed further light on the biogeographical pattern of Acronychia. The inclusion of several additional taxa in these datasets revealed that the separation of the New Guinean and Australian species is not as strict as the analyses of the concatenated matrix suggests, since the New Guinean species A. emarginata and A. rugosa are found in an Australian clade and one accession of A. acronychioides is part of the New Guinean clade.
These first biogeographical findings should be regarded as preliminary. A more detailed analysis is only feasible in the light of a much wider taxon sampling (Appelhans et al., in preparation) since no fossil data is available for Acronychia. Using Mesquite for Ancestral Area Reconstruction can only deliver a broad overview, especially since the method is unable to take into account phylogenetic uncertainty, putative extinction events, incomplete taxon sampling, non-random distribution of missing taxa and varying dispersal probabilities through time (e.g. caused by tectonic movement, sea-level changes).

Phylogeny and Systematics
Our study reveals that Acronychia is monophyletic only if Maclurodendron is included. This is in agreement with the study of Appelhans et al. [1], which also included one Maclurodendron, but far fewer Acronychia species than the present study. Further, the results support previous findings [1] that Melicope is paraphyletic with respect to Acronychia and other genera. Appelhans et al. [1] found that Acronychia was nested within Melicope, but support for this topology was only strong in Bayesian analyses and not in ML analyses, and only a few species of Acronychia were sampled. Integration of our dataset with the Melicope dataset of Appelhans et al. [1] will shed further light on the relationship of the two genera (Appelhans et al., in preparation). Within Acronychia, Maclurodendron is most closely related to the widespread A. pedunculata (Fig 1). The two taxa have a largely congruent distribution. Maclurodendron ranges from Sumatra, Malay Peninsula, Borneo, and the Philippines to Vietnam, Hainan and Guangdong (China), and A. pedunculata ranges throughout mainland southern Asia and Malesia [1,3,32,57]. Maclurodendron and A. pedunculata are morphologically similar in having completely syncarpous fruits and unifoliolate leaves, characters that are variable in Acronychia. Differences between the two taxa include unisexual vs. bisexual flowers, narrowly imbricate vs. valvate petals and differences in indumentum and shape of staminal filaments [32].
Since Maclurodendron is deeply nested within Acronychia and all Acronychia species have bisexual flowers, the unisexual flowers of Maclurodendron are interpreted as derived. Only one species of Maclurodendron was sampled in this study and it is therefore not clear if dioecy evolved several times from an ancestral bisexual population or if Maclurodendron is monophyletic within Acronychia and dioecy evolved only once. The differences among Maclurodendron species are minute and consist of differences in indumentum, sizes of floral organs, leaf texture and ornamentation of the seed coat [32]. Of these, the ornamentation of the seed coat has been regarded as taxonomically informative [32]. The seed coat of Acronychia has been described as smoothish to finely tuberculate, muricate or rugose ( [3] p. 472), but there is no detailed description of which Acronychia species exhibit which ornamentation. The morphological differences among Maclurodendron species are minor compared to those between Maclurodendron and its closest relative A. pedunculata. Therefore we hypothesize a single origin of dioecy in Maclurodendron, to be tested with more detailed phylogenetic analysis.
The results indicate that Maclurodendron should be synonymized. If included in Acronychia, three new combinations are required (three of the six species have names available in Acronychia). However, a combined study of Acronychia and Melicope currently underway (Appelhans et al., in preparation) may support different generic circumscriptions in this group, therefore it would be premature to make nomenclatural changes here. Further analysis including samples of other species of Maclurodendron would be desirable to confirm this placement.
Three samples of A. littoralis were included in our analyses, with one sample placed in Clade 2 and the other two samples placed in Clade 4 (Fig 1). Acronychia littoralis consists of two different morphological and geographical types each of separate hybrid origin [50]. The parents of one type are A. imperforata and A. wilcoxiana, while the parents of the second type are A. imperforata and A. oblongifolia (which is a close relative of A. wilcoxiana) [50]. The hybrid nature of this species can be construed from the fruit morphology. Acronychia littoralis shares the woody mesocarp with A. imperforata and the other members of Clade 4, and subapocarpous fruits with A. oblongifolia and A. wilcoxiana (Clade 2, Fig 1). The ITS and ETS sequences proved to be by far the most variable in our study as compared to NIAi3, trnL-trnF and psbA-trnH and therefore probably have the biggest influence on tree topology. ITS and ETS were not cloned in this study and maternal haplotypes were probably sequenced for one accession and paternal haplotypes for the other, explaining why one accession of A. littoralis clusters with A. wilcoxiana and A. oblongifolia, and the other with Clade 4 (Fig 1). We performed a separate phylogenetic reconstruction excluding the A. littoralis samples in order to evaluate if the inclusion of this hybrid species has an effect on tree topology and support. Both tree topology and support values remained the same except for the relationships within Clade 2 (Fig 1; results of analysis without A. littoralis not shown).

Fruit Evolution and Hartley´s Evolutionary Lines
Hartley [3] subdivided Acronychia into seven groups based on the presence or absence of septicidal fissures in the fruit, the length of the septicidal fissures and the structure of the mesocarp (fleshy and not differentiated from the exocarp; spongy-crustaceous; (sub)woody). He considered fruits with septicidal fissures and a fleshy epicarp (= exocarp + mesocarp) as the 'primitive condition' and fruits without septicidal fissures and spongy-crustaceous mesocarp and especially those with woody mesocarp as derived [3]. From these groupings, Hartley [3] inferred four evolutionary lines, all of which contain one or two species with the 'primitive' fruit characteristics. He assumed that the evolution towards the more 'advanced' fruit characters happened in parallel in these independent groups. Our results show that neither the fruit morphological 'groups' nor the 'evolutionary lines' are monophyletic (Figs 1 and 4). Despite this, fruit morphological characters are of phylogenetic and taxonomic significance in Acronychia. The three most early-branching clades of the Acronychia phylogeny (clades 1 to 3 in Figs 1 and 4) contain species with septicidal fissures only. Acronychia octandra, which is sister to the rest of the genus, is of special interest for the evolution of fruit morphology in Acronychia. This species has almost completely apocarpous fruits and differs from all other Acronychia in having a stylar beak [11]. Furthermore the exo-and mesocarp of this species is thin and chartaceous [4,11] therefore the fruit of A. octandra may be regarded as intermediate between the capsular or follicular fruits of Melicope and other close relatives, and the subapocarpous or syncarpous drupes of the other Acronychia species. The fact that Clades 2 and 3 (Fig 1) consist of species with septicidal fissures (= subapocarpous fruits) further supports the hypothesis that syncarpous fruits in Acronychia are derived from an apocarpous ancestor. These results are in agreement with previous hypotheses that fleshy drupes and syncarpous gynoecia are derived character states for the Melicope-Acronychia group as well as for other Australasian Rutaceae ( [2,3,12,13] and literature cited in [2]). Completely syncarpous fruits have evolved several times within Acronychia. Clade 4 (Fig 1) contains several species with syncarpous fruits that are found in three subclades. The sampled New Guinean species and the widespread species have fully syncarpous fruits [3]. It is worth noting that while most of the unsampled species from New Guinea also have fully syncarpous fruits some do not, e.g. A. goniocarpa and A. montana.
It is striking that the Acronychia species with the largest distributions all have fully syncarpous fruits (without septicidal fissures) and usually a woody or subwoody mesocarp. These species include A. pedunculata and A. trifoliolata (var. trifoliolata), the only species occurring west of Australasia, and the Australian species A. acronychioides and A. imperforata [3,4]. After A. acronychioides and A. imperforata, A. oblongifolia and A. laevis (which also occurs in New Caledonia), are the most widespread species in Australia. Acronychia laevis has fully syncarpous fruits, but its mesocarp is not differentiated from the fleshy exocarp [3,4] and A. oblongifolia has 'inconspicuous septicidal fissures extending for not more than half of the length' of the fruit ( [3] p. 496). In contrast, Australian species with septicidal fissures all have narrower distributions. It should be noted that not all Australian and New Guinean species with syncarpous fruits (and no septicidal fissures) are widespread: A. aberrans of north-east Queensland and A. kaindiensis of New Guinea, for example, are narrow endemics. It would appear that though having syncarpous fruits does not mean a species will be widely distributed it may predispose a taxon to being dispersed more easily. A possible explanation could be that fruits with deep septicidal fissures fall apart when they are eaten by birds, so that only some seeds are swallowed, and/or the syncarpous fruit and the (sub)woody mesocarp serves as a protective layer for the seeds. These two hypotheses need to be tested by field observations of feeding behavior and germination experiments.

Biogeographical insights
Mapping biogeographical areas onto the phylogenetic tree from the concatenated matrix shows that the early branching lineages within Acronychia exclusively contain Australian taxa and that the New Guinean and widespread species form a monophyletic group placed in a derived position within Acronychia (Figs 1 and 5). A geographic origin of the Acronychia clade in Australia is inferred and a single colonization event either via (stepping-stone) dispersal or range expansion during times of land bridges between Australia and New Guinea might have brought Acronychia to New Guinea. Two New Guinean species (A. emarginata and A. rugosa) were sampled in the separate ETS analyses only (Fig 2) and were placed in an Australian subclade. This suggests that the historical biogeography of Acronychia with respect to Australia and New Guinea is more complex. However, the lack of resolution in the ETS tree means that choosing among alternative explanatory hypotheses (e.g. vicariance, or multiple dispersals) must await further resolution of relationships.
The clear separation of Australian species from the New Guinean and widespread species is surprising given the geology of the Sahul shelf, which contains New Guinea, Australia, and nearby islands. Today, New Guinea and Australia's Cape York Peninsula are separated by the Torres Strait, which is about 150 km wide and contains hundreds of islands, reefs and shoals [60]. Most parts of the Torres Strait are only a few meters deep and New Guinea and Australia have been connected by land bridges during glacial periods throughout the Quaternary [60,61,62]. Given the short distance, the many islands that could have served as stepping-stones for dispersal, the land bridges during the Quaternary, and the availability of birds as dispersal vectors, one would not expect such a clear distinction between New Guinean and Australian Acronychia species. On the other hand, Acronychia is one of many taxa for which the Torres Strait forms a relatively strict barrier, regarded by Van Steenis ( [63] p. 72) as 'one of the main demarcations of the Palaeotropic plant world'. Differences in ecology and vegetation types have been hypothesized as being the main factors of this demarcation, rather than geography [60,61,64,65,66].
New Guinean and Australian species of Acronychia occur in different vegetation types. Australian species occur in rain forests mostly from sea level to about 1000 m elevation though A. chooreechillum occurs in montane rain forest and shrubbery up to 1600 m elevation [4]. The majority of the New Guinean species occur in montane rain forests or cloud forests from 1500 m to 2500 m, and several species also occur in subalpine forests and shrubberies up to 3260 m [3,5,9]. Only four New Guinean endemic Acronychia species are found at elevations below 1500 m. Three of them, A. dimorphocalyx, A. gurukorensis and A. reticulata occur on the northern side of the New Guinean Highlands [3] and are thus geographically separated from the Australian species. The fourth species occurring at low elevations, A. normanbiensis, is endemic to Normanby Island in the south-east of Papua New Guinea and Hartley [3] considered this species to be closely related to A. kainandiensis, a montane species from central New Guinea. Of these four species only A. reticulata has been included in this study and it is deeply nested within the New Guinean clade. The inclusion of other species, especially A. normanbiensis, in future studies would be highly desirable.
Acronychia is known, at least in Australia, to be dispersed by birds including bowerbirds (green catbird, regent bowerbird, satin bowerbird), currawongs (pied currawong) and pigeons (topknot pigeon, wompoo fruit-dove, rose-crowned fruit-dove) [67,68]. Of these, only the two fruit-doves also occur in New Guinea [69,70]. The rose-crowned fruit-dove has been reported as a vagrant species in the Fly River delta region at sea-level. The wompoo fruit-dove is widespread in New Guinea and travels between New Guinea and Australia, but it is restricted to lowland forests from sea level to 1400 m in New Guinea [70]. The diet of the wompoo fruit-dove in New Guinea has been studied and no evidence that it eats Acronychia fruits has been reported [71]. Dependency on birds that do not cross the Torres Strait could be an additional factor explaining the strict phylogenetic separation of New Guinean and Australian Acronychia species.
Most New Guinean species of Acronychia have completely syncarpous fruits and only six species exhibit apical fissures like most Australian species [3,5]. Of these six species, four are known only from the type collection [3,5] and only one of them-A. rugosa-was included in this study. This species, together with another New Guinean endemic, A. emarginata, was found to belong to an Australian subclade (Clade 4 in Fig 1) in the ETS analyses (Fig 2) but this relationship was only supported in the ML analysis. The lack of resolution within this subclade impedes the determination of the closest relatives of these two species.
The placement of the New Guinean A. emarginata within the otherwise Australian Clade 4 is highly surprising given that it is a montane species occurring between 1760-2370 m elevation, and that it is morphologically similar to A. murina [3], a member of the New Guinean clade (Fig 1). Additional data are needed to ascertain the position of this species as well as A. rugosa and the other New Guinean species with apically fissured fruits. The position of A. emarginata in the Australian clade and the putative close relationship of the New Guinean species suggest that Torres Strait is a significant but not absolute, geographical demarcation line separating Acronychia lineages.
Hartley [3] hypothesized an extra-Australian origin of the Australian species A. acronychioides and A. imperforata based on an assumed close relationship with the widespread A. trifoliolata and A. pedunculata respectively. These relationships are not substantiated by our data (Fig 1) as these two Australian species are deeply nested within the Australian Clade 4 (Fig 1).
Only three Acronychia species, and Maclurodendron (six spp.), occur outside of New Guinea and Australia. Acronychia laevis from Australia and New Caledonia is a member of the Australian Clade 4 (Fig 1). Acronychia pedunculata, and to a lesser extent Maclurodendron range from mainland South-East Asia to New Guinea, whereas A. trifoliolata is found from Java and Christmas Island to the Solomon Islands [3,32]. The relatively large geographical ranges of A. pedunculata and A. trifoliolata coincide with a wide variety of ecological niches occupied by these species. Both species occur from near sea level to montane elevations (up to 2200 m in A. pedunculata and 2400 m in A. trifoliolata) and a variety of vegetation types including coastal scrubs, primary and secondary rain forests, monsoon forests and montane rain forests [3]. Maclurodendron contains several species with narrow distributions at low elevations, and one species, M. porteri, with a larger distribution ranging from the Malay Peninsula and Sumatra to the Philippines [32]. Maclurodendron porteri is found over a relatively wide altitudinal range (from sea level to up to 1500 m) and in a diversity of vegetation types (primary and secondary rain forest, heath forest) indicating that the species has a wide ecological niche like A. pedunculata and A. trifoliolata.
Acronychia pedunculata and Maclurodendron form the sister clade to all other New Guinean species including A. trifoliolata (Fig 1), so that it is not clear if the A. pedunculata clade originated in Australia or if the clade split from the ancestor of the New Guinean clade after New Guinea was colonized. Acronychia trifoliolata is deeply nested within the New Guinean clade (Fig 1), sister to the two montane species A. ledermannii and A. pullei. Acronychia ledermannii was suggested to be a close relative of A. trifoliolata based on morphology [3]. Since several New Guinean species are missing in our study, the immediate affinities of A. trifoliolata cannot be finally addressed, but an origin of the species in New Guinea followed by range expansion and dispersal to the east (Solomon Islands) and west (to Java and Christmas Island) is likely. Since A. pedunculata and A. trifoliolata are not sisters, it is likely their distributions have resulted from two independent range expansions westwards from New Guinea which would add to the relatively few known cases of lineages of Australian origin dispersing across Wallace's Line [72].

Conclusions
This study confirms that Acronychia is monophyletic only if Maclurodendron is included. The 'evolutionary lines' within Acronychia, which are based on fruit morphological characters (connation of carpels, structure of mesocarp; [3]), are not monophyletic and should not be used to define subgenera. The early-branching clades within Acronychia consist of apocarpous (A. octandra) or subapocarpous species and syncarpous fruits evolved from this condition. Maclurodendron is most closely related to A. pedunculata. Both taxa have almost congruent distributional ranges and are morphologically similar. The dioecious flowers of Maclurodendron are most likely derived from a bisexual ancestor, but since only one out of six species of Maclurodendron was included in this study, it is not clear if dioecy in Maclurodendron has one or several origins. A strict separation of the Australian from the New Guinean and extra-Australasian species was inferred. Since the species north of the Torres Strait occur in different habitats compared to the Australian species, the strict separation probably has ecological rather than geographical/geological causes. The geographical origin of Acronychia is in Australia. Only two species of Acronychia occur outside of Australasia. Both species are part of an otherwise New Guinean clade, but the two species are not immediate relatives, suggesting that the colonization of areas westward of New Guinea is the result of two independent dispersal events.