Phylogeny of Miliusa (Magnoliales: Annonaceae: Malmeoideae: Miliuseae), with descriptions of two new species from Malesia

Abstract. The molecular phylogeny of Miliusa (Annonaceae) is reconstructed, with 27 (of ca. 50) species included, using a combination of seven plastid markers (rbcL exon, trnL intron, trnL-F spacer, matK exon, ndhF exon, psbA-trnH spacer, and ycf1 exon) constituting ca. 7 kb. In addition, two new species of Miliusa are described from the Malesian area: M. butonensis sp. nov. from Buton Island, Indonesia and M. viridifl ora sp. nov. from Papua New Guinea. The former is included in the molecular phylogenetic analysis. The reconstructed phylogeny corresponds well to the informal morphological grouping proposed earlier. A revised key to 13 Austro-Malesian species of Miliusa is provided.


Introduction
The genus Miliusa Lesch.ex A.DC. (de Candolle 1832) (Annonaceae) comprises approximately 50 species of shrubs, or small to large trees, distributed from the Indian subcontinent, southern China and mainland Southeast Asia to Southeast Asian islands, New Guinea (including D'Entrecasteaux Islands and Louisiade Archipelago) and northern Australia (Chaowasku & Keßler 2006).It belongs to the tribe Miliuseae of the subfamily Malmeoideae (Chatrou et al. 2012).Members of the tribe Miliuseae are almost exclusively Asian (including New Guinea, Australia, and the western Pacific islands).Only two clades within this tribe consist of non-Asian members: one clade of four Neotropical genera and another clade of Afro-Madagascan species which are part of the recently described genus sister to Miliusa: Hubera Chaowasku (Chaowasku et al. 2012).
According to Chaowasku & Keßler (2006), Miliusa is circumscribed by having (1) equally-sized sepals and outer petals, both of which are much smaller than the inner petals; (2) a densely hairy torus; (3) miliusoid stamens (sensu Mols & Keßler 2003a), i.e. stamens that are loosely arranged without conspicuously dilated connective tissue covering the thecae; and (4) four-part-lamellate ruminations of the endosperm.Ten species were recognized according to the revision of the genus in the Austro-Malesian area (Mols & Keßler 2003b).One additional species, M. lanceolata Chaowasku & Kessler (Chaowasku & Keßler 2006), was later described from D'Entrecasteaux Islands and Louisiade Archipelago, southeast off Papua New Guinea.New Miliusa species from southwestern India (Narayanan et al. 2010(Narayanan et al. , 2012) ) as well as the Indian eastern Himalaya (Chaowasku 2013) have recently been described.Further, seven new species from Thailand are being added to this medium-sized genus by Chaowasku & Keßler (in press) who use floral/inflorescence morphology to elaborate the four informal groups within Miliusa first introduced by Chaowasku & Keßler (2006).These four morphological groups were the starting point to systematically study this genus further.In order to obtain additional evidence supporting the four mentioned groups, their pollen was investigated in detail (Chaowasku et al. 2008).The aim of the present study is to test whether each of the four morphological groups of Miliusa is monophyletic by means of molecular phylogenetic analysis.
In the course of selecting specimens for DNA extraction, we came across two collections, one from Buton Island (Indonesia) and another from Papua New Guinea, which are different from other described species in the Austro-Malesian area.After a thorough examination and comparison with similar species, we have concluded that they both represent undescribed species of Miliusa, which are herein formally described as M. butonensis sp.nov.and M. viridiflora sp.nov.The former is included in the molecular phylogenetic analysis.The number of Miliusa species in the Austro-Malesian area is thus increased to 13 and a key to these species is provided.

Molecular phylogenetic analysis
All accessions belong to the subfamily Malmeoideae (Appendix).Twenty-seven accessions of Miliusa covering the entire morphological variation known comprise the ingroup.The outgroups consist of 14 accessions, 11 of which were included as representatives of related genera in the tribe Miliuseae.Seven plastid markers, i.e. rbcL exon, trnL intron, trnL-F spacer, matK exon, ndhF exon, psbA-trnH spacer, and ycf1 exon, were amplified (see Table 1 for the number of included, variable, and parsimony informative characters).In total, 7033 characters, including six separately coded indels were included in the analyses.Indel coding follows Simmons & Ochoterena (2000).The reverse complement of 15 continuous nucleotides in the psbA-trnH marker for roughly half of the accessions sequenced was present and altered into the reverse complement, following Pirie et al. (2006).
All methods of DNA extraction, amplification, and sequencing performed in Chaowasku et al. (2012) were used in the present study.Due to a poor quality of the extracted DNA or unavailability of leaf material, we could not produce seven markers for all accessions (see Appendix, Table 1).Sequences were edited using the program Staden version 1.7.0 (http://staden.sourceforge.net/)and subsequently manually aligned.Some sequences were obtained from previous studies (Mols et al. 2004a(Mols et al. , 2004b;;Pirie et al. 2006;Chaowasku et al. 2012).Maximum parsimony analyses were performed in TNT version 1.1 (Goloboff et al. 2008).All characters were equally weighted and unordered.Multiple most parsimonious trees were generated by a heuristic search of the combined data, with 6 000 replicates of random sequence additions, saving 10 trees per replicate, and using tree bisection and reconnection (TBR) branch swapping algorithm.Clade support was measured by symmetric resampling (SR), which is not affected by a distortion (resulting in incorrectly estimated percentages) as with some bootstrap and jackknife methods (Goloboff et al. 2003).A default change probability was used.Four hundred thousand replicates were run, each with two replicates of random sequence additions, saving one tree per replicate.Groups with SR of ≥ 85%, 70-84%, and ≤ 69% were considered strongly, moderately, and weakly supported, respectively.
Bayesian analysis was performed in MrBayes version 3.1.2(Ronquist & Huelsenbeck 2003).Two independent runs were simultaneously run; each run comprised four Markov-chain-Monte-Carlo (MCMC) chains and was set for 10 7 generations.The data matrix was divided into seven partitions [trnL intron and trnL-F spacer were included in the same partition (= trnLF)], including a set of binary indel coding.The most appropriate model of sequence evolution for each partition was selected by Akaike information criterion (AIC) scores, using FindModel (http://www.hiv.lanl.gov/content/sequence/findmodel/findmodel.html).The default model as well as the command 'coding=variable' were applied for the binary indel partition.The default prior settings were used except for the ratepr [=variable] and brlenspr [=unconstrained:exp(100)]. The latter prior setting was used to prevent the MCMC chains from being trapped in the areas of parameter space with unrealistically high values for the tree length parameter, resulting in a false convergence or a failure to reach convergence after hundreds of millions of generations (Marshall 2010).The temperature parameter was set to 0.1.Trees and all parameter values were sampled every 1000 th generation.Convergence of the runs was checked by both the standard deviation of split frequencies and the values for effective sample sizes (ESS) using Tracer version 1.5 (Rambaut & Drummond 2009).The 50% majority-rule consensus tree was generated from the two runs combined, with 10% of the first trees removed as burn-in.Groups with posterior probabilities (PP) of ≥ 0.96, 0.91-0.95,and ≤ 0.90 were considered strongly, moderately, and weakly supported, respectively.

Taxonomy/morphology
Measurements/observations of the new species were made from herbarium specimens of A, BRI, CANB, E, K, L, U herbaria.The indumentum terminology follows Hewson (1988).The term 'velvety' is equivalent to densely hairy/with dense hairs, whereas 'puberulous' is equivalent to sparsely hairy/ with sparse hairs.Morphological data of the 11 known species of Miliusa in the Austro-Malesian area were from Mols & Keßler (2003b), Chaowasku & Keßler (2006), and personal observations (= all specimens cited in Mols & Keßler 2003b).Morphological data of other species included in the molecular phylogenetic analysis were from Chaowasku (2013), Chaowasku & Keßler (in press), and personal observations (= voucher specimens for molecular phylogenetic analysis plus a few more specimens, see Appendix).The term 'glandular structures' in the present study means that such structures look like glands but further anatomical confirmation is required.When only a single measurement/observation was made, the word 'circa (ca.)' was added.

Molecular phylogenetic analysis
The maximum parsimony analysis of combined data sets resulted in 135 most parsimonious trees with 980 steps (results not shown).The consistency and retention indices (CI, RI) were 0.86 and 0.88,

DNA region
No  respectively.For Bayesian analysis, the substitution model was generalized time-reversible plus gamma (GTR + G) for all partitions except the trnLF partition (= trnL intron + trnL-F spacer), which had Tamura-Nei plus gamma (TrN + G) model.The final standard deviation of split frequencies was < 0.002 and all ESS values after discarding the burn-in were > 1200, both indicating convergence of the runs.

Etymology
Named after the Buton Island (Indonesia) where this species is endemic.

Distribution, habitat and phenology
Indonesia (Buton Island, Fig. 4), occurring in forests on flat ridge-tops, with slightly broken canopy due to rocky terrain of raised coralline limestone.Elevation: ca.300 m.Fruiting: November.

Field notes
Branches horizontal and foliage in flat sprays.Bark grey-brown, ca. 4 mm thick overall, ± smooth with fine vertical cracks and rows of low lenticels, wood straw-colored, cut bark and wood pleasantly aromatic.Young leaves ± yellow-green, mature leaves mid green on both sides.

Etymology
The epithet refers to the light green flowers (probably also at anthesis).

Field notes
A small tree.Leaves mid green.Flowers light green.

Notes
In some couplets differences are small, therefore when there is any ambiguity, it is advised to consult the descriptions of relevant species.) each does not form a clade; clades A and D each consists of members of both groups (see Fig. 1).The M. velutina group seems to have been characterized principally by a likely symplesiomorphy: an absence of conspicuous glandular structures inside the inner petals (Fig. 6B, K), whereas the main feature characterizing the M. horsfieldii group [inner petals with narrow glandular structures running inside along their bilateral midline (Fig. 6A, J, L)] seems to have evolved multiple times in Miliusa.
Detailed ancestral character reconstructions in combination with a denser taxon sampling, however, are needed before any solid conclusion on character evolution occurred in Miliusa can be drawn.
Besides the inner petal morphology, flower and/or inflorescence position also corresponds to the phylogenetic results, i.e. all species recovered in clade C and most species recovered in clade B possess After a thorough examination on the floral morphology of Miliusa velutina, a member of clade A (Fig. 1), peculiar structures have been observed.At the base inside the inner petals, there are thickened structures (Fig. 6C) hidden at female anthesis, but as male anthesis begins and continues, these structures are gradually becoming exposed.(Chaowasku & Keßler 2006)].These differences convince us to recognize the former as new species.
So far neither macromorphological nor pollen morphological (see Chaowasku et al. 2008) features have been found to be able to distinguish species of clade A from those of clade D. Nevertheless, Miliusa viridiflora sp.nov. is more likely to belong to clade D because (1) clade A thus far known contains only continental Asian species and (2) all species known to occur in New Guinea belong to clade D (Fig. 1).The peduncles plus rachis (if present) plus pedicels of M. viridiflora sp.nov.(Fig. 5a, b) and of M. lanceolata (Chaowasku & Keßler 2006) are notably long.This trait may be associated with bat dispersal syndrome since the fruits eventually set will be clearly separated from the foliage, and hence can be more easily detected (Marshall 1983).In Miliuseae, this feature has been earlier reported in some New Guinean species of Pseuduvaria Miq.(Miquel 1858) (Su & Saunders 2006;Su et al. 2008).
It is worth conducting an anatomical investigation of the inner side of the inner petals of Miliusa to reveal the ontogeny and possible functions of the glandular [and seemingly non-glandular (e.g.Fig. 6C,  E)] structures.Probably these differentiations are associated with the different pollination strategies.So far there has been no detailed study on pollination biology of Miliusa; however, according to Mols & Keßler (2003b), fruit flies were observed to visit the flowers of M. horsfieldii.A detailed pollination biological study is required to determine if this kind of insect is potential pollinators for this species.
Certain species of clade B exhibit ± transparent, window-like structures at the base of the inner petals (Chaowasku & Keßler in press), e.g.M. campanulata and M. thorelii Finet & Gagnep. (Finet & Gagnepain 1907; see Fig. 6E).Flies are likely to be the potential pollinators for these species as they are lured to crawl inside to the stamens/stigmas by light (Dafni 1984).
There are correlations between clades and habitat preferences in Miliusa.Most species (expected to be part) of clades A and D prefer drier habitats (e.g.deciduous/dipterocarp forests), resulting in a various degree of deciduous lifecycle exhibited by a number of species (Mols & Keßler 2003b).In contrast, the majority of the species (expected to be part) of clades B and C prefer more humid habitats, e.g.dry/moist evergreen forests (Mols & Keßler 2003b;Chaowasku & Keßler in press).It is apparent that habitat shifts occurred in Miliusa; these shifts might correspond to the paleoclimate.
More species of Miliusa, especially the Indian, Philippine, and Vietnamese ones, need to be included in order to reconstruct a more robust molecular phylogeny, which will be the ground for the study in, for example, biogeography/molecular dating and character evolution.

Figures 1 (
Figures 1 (cladogram) and 2 (phylogram) show 50% majority-rule consensus trees derived from the Bayesian analysis, with support values indicated in Fig. 1.Miliusa is monophyletic with maximum support (SR 100%; PP 1.00).It is sister to Hubera with strong support (SR 90%; PP 0.98).Within Miliusa, four strongly supported (SR ≥ 85%; PP ≥ 0.96) clades have been identified (Fig. 1: clades A, B, C, D).Clade A is sister to clade B whereas clade C is sister to clade D, both with strong support.A clade comprising clades A and B is sister to a clade consisting of clades C and D. Taxonomy

Fig. 2 .
Fig. 2. 50% Bayesian majority-rule consensus phylogram of combined seven plastid markers, showing branch length proportional to amount of lineage sequence divergence.Scale bar unit: substitution per site.

Table 1 .
Important descriptive values of sequence data.NA = not applicable.
These structures seem to be non-glandular; however, anatomical comparisons with the glandular structures observed in the species of clade C (Figs 3c; 6G-I) are likely to shed light on whether they are homologous.Miliusa butonensis sp.nov. is only known from the type specimens.Results of the molecular phylogenetic analysis, nonetheless, assure the new species status.Miliusa butonensis sp.nov. is sister to a clade comprising three mainland Asian species (Fig.1).These three species share one remarkable feature: (sub-)cordate leaf base, whereas M. butonensis sp.nov., and the remaining species of clade C, M. fusca Pierre (Pierre 1881) and M. fragransChaowasku & Kessler (Chaowasku & Keßler in press), do not.
Sawasdee et al. 20103bSawasdee et al. , 2013a) )s unexpected since all other species (expected to be part) of clade C thus far known occur on mainland Asia only.Phytochemically, it is worthwhile to note that neolignans have been found as principal secondary metabolites in two species of clade C: M. mollis(Sawasdee et al. 2010(Sawasdee et al. , 2013a) )and M. fragrans(Sawasdee et al. 2013b), but have not been reported to occur in any species (expected to be part) of clade A, B or D so far investigated (seeSawasdee et al. 2010).If it is eventually proved that neolignans really occur only in the species (expected to be part) of clade C, including M. butonensis sp.nov., this class of natural product could be developed as a chemotaxonomic marker.Miliusa viridiflora sp.nov. is also only known from the type specimens; however, it is readily distinguishable from its most morphologically similar species, M. lanceolota, by both vegetative and generative features (see diagnosis).Additionally, the elevation where both species occur is considerably different[ca.1220 m in M. viridiflora sp.nov. vs. 2-20 m in M. lanceolata