Ancestral range reconstruction of remote oceanic island species of Plantago (Plantaginaceae) reveals differing scales and modes of dispersal

Abstract Aim The aim of this study was to resolve the phylogenetic placement of island taxa, reconstruct ancestral origins and resolve competing hypotheses of dispersal patterns and biogeographical histories for oceanic island endemic taxa within subgenus Plantago (Plantaginaceae). Location Juan Fernández Islands, the Auckland Islands, Lord Howe Island, New Amsterdam Island, New Zealand, Tasmania, Falkland Islands, Rapa Iti and the Hawaiian Islands. Taxon Island endemics within Plantago (Plantaginaceae), a globally distributed taxonomic group comprising approximately 250 species. Methods We use Bayesian phylogenetic and divergence time analyses and historical biogeographical analysis of molecular sequence data to infer the ancestral origins of the oceanic island species in Plantago. Results Taxa within subgenus Plantago form clades based on geographic proximities and challenge previous phylogenetic relationships and classification based on morphology. We infer that biogeographic histories of oceanic island taxa from multiple islands were shaped by dispersal at different scales and possibly by different types of birds. The highly remote Hawaiian Islands and Rapa Iti were colonized from North American taxa in a pattern corresponding to known migration routes of large marine birds, rather than from New Zealand as previously hypothesized. The island endemics of Juan Fernández, the Falkland Islands, Lord Howe, Auckland Islands and New Zealand are found to have sources in the nearest continental areas. The analyses confirm recent speciation within subgenus Plantago – which is particularly heightened in island lineages in Hawaii and Rapa Iti – but show slightly older divergence times than previous molecular dating studies. Main conclusions Using molecular data to infer ancestral ranges for plants with uncertain taxonomic relationships can greatly improve our understanding of biogeographical histories and help elucidate origins, dispersal modes and routes in widespread lineages with complex distribution patterns such as Plantago. We improve understanding of important floristic exchange areas between continents and islands as a result of long‐distance dispersal. We infer that a combination of both stepping stone dispersal and extreme long‐distance dispersal can shape insular floras, and that multiple floristic areas can be the sources of closely related island taxa. However, despite the successful dispersal of Plantago, radiation in island archipelagos is generally limited suggesting specific traits may limit diversification.

as a result of long-distance dispersal. We infer that a combination of both stepping stone dispersal and extreme long-distance dispersal can shape insular floras, and that multiple floristic areas can be the sources of closely related island taxa. However, despite the successful dispersal of Plantago, radiation in island archipelagos is generally limited suggesting specific traits may limit diversification.

K E Y W O R D S
biogeographical range reconstruction, bird dispersal, island taxa, long-distance dispersal, oceanic islands, Plantago
In the case of plants, the movement of propagules either internally or externally by birds is recognized as the most common type of LDD responsible for the distribution of disjunct plant lineages on remote oceanic islands, though dispersal by oceanic drift and weather events also play roles (Carlquist, 1966(Carlquist, , 1967Gillespie et al., 2012;Kistler et al., 2014;Nathan et al., 2008;Pole, 1994;Sanmartin & Ronquist, 2004). Birds are more often implicated as the main dispersal vectors for island plants based on an over-representation of plant traits that favour bird dispersal in island taxa (Baldwin & Wagner, 2010;Carlquist, 1967;Gillespie et al., 2012). The nearest landmasses are often implicated as the primary sources for island taxa, yet, as a result of bird-mediated LDD across extreme distances such as via the Pacific or Asia-Australian bird flyways, thousands of kilometres can separate some of the most closely related taxa (Baldwin & Wagner, 2010;Gillespie et al., 2012). Thus, LDD events mediated by bird vectors can result in multiple theories of biogeographic origins for island plant lineages (Birch & Keeley, 2013;de Queiroz, 2005;Kainulainen, Razafimandimbison, Wikstrom, & Bremer, 2017; le Roux et al., 2014). Applying biogeographical models and using molecular phylogenetic relationships to reconstruct source areas for island taxa can help resolve complex biogeographical histories, identify important areas of floristic exchange and elucidate dispersal routes that shape island floras (Bacon, Simmons, Archer, Zhao, & Andriantiana, 2016;Bouckaert et al., 2014;Ho et al., 2015;Johnson, Clark, Wagner, & McDade, 2017;Matzke, 2014).
Plantago L., a genus of approximately 250 species, is a model group to study LDD processes and compare different hypotheses regarding dispersal modes and routes, due to its worldwide distribution, high dispersal capabilities and high number of single island endemic taxa (Dunbar-Co, Wieczorek, & Morden, 2008;Hassemer, De Giovanni, & Trevisan, 2016;Rahn, 1996;Tay, Meudt, Garnock-Jones, & Ritchie, 2010a). The genus is largely temperate in distribution, but occurs at high altitudes in tropical areas and on oceanic islands (van der Aart & Vulto, 1992). Previous phylogenetic and molecular divergence analyses have estimated that Plantago s.l. (including Littorella P.J.Bergius) diverged from its closest known relative, Aragoa Kunth, in the late Miocene to Pliocene, 7.1 million years ago (Ma) (Bello, Chase, Olmstead, Rønsted, & Albach, 2002; or 2.8 Ma (Tay et al., 2010a). This finding of recent diversification of the genus rules out earlier hypotheses of vicariance in explaining the global distribution of the genus (i.e. as proposed by Rahn, 1996), and thus, long-distance dispersalpresumably by birdsis the accepted scenario for Plantago (Dunbar-Co et al., 2008;Rønsted et al., 2002;Tay et al., 2010a). Not only are the seeds of Plantago species known to be eaten by birds and other animals (Buse & Filser, 2014;Czarnecka & Kitowski, 2013;Panter & Dolman, 2012) but also Plantago species often grow alongside graminoids and other plants that are known to be typically eaten as fodder by birds prior to long migratory flights (Carlquist, 1966;Meudt, 2012;Rahn, 1996). Furthermore, dispersal in the genus is thought to be facilitated by the mucilaginous properties of wetted seeds, a specialized adaptation that assists with seed dispersal by increasing chance of seeds adhering to animals (Fischer et al., 2004;Rønsted et al., 2002;Tay et al., 2010a) or by attracting animals and facilitating dispersal via ingestion (Buse & Filser, 2014;Western, 2012). The mucilage in P. lanceolata L. seeds for example has been shown to be strongly adhesive to feathers and fur, especially when it has dried, which increases the potential of being transported externally over vast distances (Kreitschitz, Kovalev, & Gorb, 2016), while mucilage in P. major L. has been shown to attract invertebrates and facilitate endozoochory (Buse & Filser, 2014). IWANYCKI AHLSTRAND ET AL.

| 707
In addition to its adaptations to long-distance dispersal, the majority of Plantago species have traits that likely assist in the successful colonization of new locations, including remote oceanic islands. For example, species are wind pollinated and many are self-compatibletwo important traits that are common in other island endemics (Carlquist, 1967;Rahn, 1996;Stuessy, Crawford, & Ruiz, 2018). Additionally, many Plantago island endemics grow on cliffs and other inaccessible habitats that may serve as refuges from natural threats, thereby facilitating successful establishment and survival in new areas (Dunbar-Co, Sporck, & Sack, 2009;Stuessy et al., 2018).
The genus Plantago has been classified by Rahn (1996)  Albicans, for example, has subsequently been included in subgenus Psyllium based on molecular phylogenetic studies , and subgenus Littorella was found to be sister to the remainder of Plantago and is most often considered a separate genus (Hassemer, Moroni, & O'Leary, 2018;Hoggard, Kores, Molvray, Hoggard, & Broughton, 2003;Kolář, 2014).
Subgenus Plantago, sensu Rahn (1996), is the largest of the subgenera with about 130 species currently recognized (Rahn, 1996;Rønsted et al., 2002). The subgenus is monophyletic (Hoggard et al., 2003;Rønsted et al., 2002;Tay et al., 2010a) and is distributed on all continents. This group is the focus of this work, as the subgenus includes the highest number of native Plantago species on oceanic islands, yet taxonomic, phylogenetic and biogeographic relationships between taxa in the subgenus remain some of the most poorly resolved, and morphological variation between many of the species is low (Meudt, 2012;Rahn, 1996;Rønsted et al., 2002). Previous molecular analyses have found Rahn's (1996) taxonomic sections within subgenus Plantago to be paraphyletic (Hoggard et al., 2003;Ishikawa, Yokoyama, Ikeda, Takabe, & Tsukaya, 2009;Tay et al., 2010a) though additional sampling is necessary to confirm this and improve our understanding of biogeographic histories. The remoteness of the oceanic islands, the extreme distance between the nearest landmasses as well as uncertainty in the taxonomic relationships have resulted in numerous and often competing hypotheses as to the biogeographical histories of the island taxa in Plantago (Dunbar-Co et al., 2008;Hoggard et al., 2003;Meudt, 2012;Rahn, 1996;Rønsted et al., 2002;Tay et al., 2010aTay et al., , 2010b. Given adaptations of Plantago plants to LDD by birds, dispersal routes could be many, and relatedness between taxa may not reflect geographic proximity. For example, some of the island taxa (P. rupicola, P. fernandezia, P. princeps s.l. and P. robusta) exhibit typical island traits such as woody stems (Carlquist, 1970;Rahn, 1996;Stuessy et al., 2018), and based on such morphological similarities, P. fernandezia is thought to be more closely related to P. princeps s.l.
in the Hawaiian Islands rather than to taxa in the Americas (Pilger, 1937;Rahn, 1996). However, the question remains whether plant propagules arrived in the Juan Fernández Islands from the western Pacific or from South Americathe closest continental landmass where extant Plantago species are known (Rahn, 1996;Stuessy et al., 2018). Along similar lines, at least two competing hypotheses exist for the species endemic to the Hawaiian Islands, either arising from a LDD colonization event from a North American ancestor or from an ancestor in Australasia via Rapa Iti (Dunbar-Co et al., 2008). subgenus Plantago were available for previous phylogenetic analyses covering seven islands or island systems with known Plantago species (most of these were restricted to Australasia [Meudt, 2012;Tay et al., 2010a], see Appendix I, Supporting Information). This severely limits the testing of biogeographic histories for island taxa in the globally distributed genus Plantago.
In this study, molecular data from 14 of the island endemics were obtainedcovering 11 of the 18 different oceanic islands or island systems where Plantago species are knownand the number of taxa represented from within subgenus Plantago was substantially increased (Appendix I, Supporting Information). We include for the first-time endemic species from Juan Fernandez Islands, Lord Howe Island, Java and Jeju. This sampling increase not only results in improved phylogenetic and biogeographical inference but also allows for a greater number of island calibration points to investigate divergence times within this challenging taxonomic group. We use phylogenetic analyses of nuclear internal transcribed spacer (ITS) sequence data along with sequence data of four plastid regions (ndhF-rpl32, rpl32-trnL, rps16 and trnLF) to produce a well-supported phylogenetic tree and further aim to (1) resolve the placement of island taxa within subgenus Plantago, (2) infer source areas for island taxa by reconstructing the most probable ancestral ranges and resolve competing hypotheses regarding the biogeographic histories of taxa and (3) infer the degree to which biogeographical proximity (nearest landmass) versus extreme long-distance dispersal events is responsible for the distribution of the taxa in a genus that is likely to be dispersed by birds across extreme long distances.

| Taxon sampling
The present study follows the most recent and comprehensive circumscription of the genus published by Rahn (1996), except for the Hawaiian taxa, in which case we follow Wagner, Herbst, and Sohmer (1990). To ascertain biogeographic histories of island taxa within subgenus Plantago, we sampled as many oceanic island taxa as possible, plus the representation of species from across the global distribution for the subgenus. In total, 11 of the 18 islands or island systems for which species from subgenus Plantago are known are represented (Appendix I). In total, 54 taxa were available for our analyses (Table 1), including 48 species belonging to subgenus Plantago (representing 37% of the subgenus) of which 14 species (and three varieties) are endemic to oceanic islands. An additional six species from other taxonomic subgenera in Plantago was included to confirm monophyly of subgenus Plantago. Initially, multiple specimens were obtained for several critical or rarely collected species, including P. aucklandica, P. canescens Adams, P. fernandezia, P. hawaiensis, P. macrocarpa Cham. & Schltdl., P. palmata Hook.f., P. princeps, P. rapensis and P. rupicola, to test DNA amplification on degraded tissue samples. In all of the above cases, the multiple specimens were monophyletic (data not shown), and therefore, only one specimen per species was used in the final analyses, except for the four varieties of the Hawaiian P. princeps s.l. (Dunbar-Co et al., 2008;Wagner et al., 1990). Material from herbarium specimens of P. robusta from St. Helena was also obtained and would have provided an additional calibration point, but could not be included due to lack of amplification of the highly degraded DNA.  Table 1. Additionally, 42 sequences from the previous study of Rønsted et al. (2002), 3 from Dunbar-Co et al.
(2008) and 2 from Hoggard et al. (2003) were downloaded from GenBank as listed in Table 1. In total, 17 species were sequenced for the first time in this study, most of the remaining included species were supplemented with additional sequence data, and 176 new sequences were submitted to GenBank.

| DNA extractions, amplification and sequencing
Total genomic DNA was extracted from 15 to 30 mg of dried leaf fragments or herbarium material following Rønsted et al. (2002).
Amplification of ITS and the trnLF intron was performed following Rønsted et al. (2002), while amplification of the intergenic spacers ndhF-rpl32 and rpl32-trnL followed Dunbar-Co et al. (2008). The rps16 intron was amplified following Oxelman, Lidén, and Berglund (1997). Primers used are listed in Appendix II. Amplified products were purified with the Qiagen PCR purification kit (Qiagen, Germany) following the manufacturer's protocols. Cycle sequencing reactions were carried out using the BigDye ™ Terminator Mix (Applied Biosystems, USA). Products were run on an ABI 3730 DNA Analyzer according to the manufacturer's protocols (Applied Biosystems, USA) at the Jodrell Laboratory in Kew Gardens, at the National Sequencing Centre, Natural History Museum of Denmark, or by Macrogen Inc. (Europe). Both strands were sequenced for each region for all taxa.

| Phylogenetic analysis
Sequences were assembled, edited and subsequently aligned with MAFFT 7.2 (Katoh & Standley, 2013) using the bioinformatics software platform GENEIOUS 9.1.8 (www.geneious.com, Kearse et al., 2012). Gaps were coded for all regions following a simple gap coding scheme (Simmons and Ochoterena (2000). The best-fit nucleotide substitution model for each marker was chosen based on the corrected Akaike information criterion (AICc) as calculated using JMODELTEST 2.1.10 (Darriba, Taboada, Doallo, & Posada, 2012). The best-fit models are listed in Appendix II (Supporting Information). The partitioned data set was analysed with MRBAYES 3.2.6 (Ronquist & Huelsenbeck, 2003), running for 5 million generations and sampling every 200 generations.
Littorella uniflora was set as the outgroup based on the results of previous phylogenetic analyses for Plantago (see Hoggard et al., 2003;Rønsted et al., 2002). Chain convergence and ESS parameters were inspected with Tracer 1.6 (Rambaut, Suchard, Xie, & Drummond, 2014) and the first 25% of the trees sampled from the posterior were discarded as burn-in. A 50% majority rule consensus tree was calculated and visualized together with the posterior probabilities using FIGTREE 1.4.3 (Rambaut, 2012). Maximum likelihood analyses were performed in RAXML (Stamatakis, 2014), defining L. uniflora as outgroup and setting the number of bootstrap iterations to 1000.

| Divergence time analysis
BEAST 2.4.7 was used to compute divergence times (Bouckaert et al., 2014). The nucleotide substitution models used in the BEAST analysis were identical to the ones used in the MRBAYES analysis (Appendix II).
Littorella uniflora was defined as the outgroup. The appropriate molecular clock model was determined by using PATHSAMPLER 1.3.4, which is integrated in BEAST 2.4.7 (Bouckaert et al., 2014). The chain length for this path sampling analysis was set at 1 million generations and the number of steps at 100. The marginal likelihood estimates for a strict, a lognormal and an exponential clock were obtained, and an uncorrelated relaxed lognormal clock was determined to be the most likely model and thus selected for the analyses.
Assuming the timing of dispersal to a datable oceanic island occurred soon after emergence of the island from the ocean, the age of the island can be used as an approximate maximum date for the IWANYCKI AHLSTRAND ET AL.

| 709
T A B L E 1 List of Plantago taxa and outgroups, including specimen information, and DNA regions used in the present study occurrence of endemic species to that island (Ho et al., 2015;Richardson et al., 2001;Rønsted et al., 2002). In the absence of reliable fossil data for Plantago, previous studies of Plantago have used the endemicity of P. stauntonii on New Amsterdam Island as calibration point (Plantago section Mesembrynia Decne.; Rønsted et al., 2002;Tay et al., 2010a). In the present study, the ages of five oceanic islands were used as calibration points for the occurrence of endemic species to those islands and a normal distribution was used for the estimation of those priors (Table 2). Using this type of data may cause overestimation of divergence times (Hipsley & Müller, 2014). We therefore applied a large confidence interval for each calibration point, allowing for the lower limit to be zero (i.e. present day). Three independent MCMC runs were performed with an uncorrelated relaxed lognormal clock prior on molecular rates and a restricted calibrated Yule speciation process prior on tree shapes. Chain length for each run was set at 100 million generations and sampling was conducted every 5,000 generations. Chain convergence and ESS parameters were inspected with TRACER 1.6 (Rambaut et al., 2014). The burn-in of each run was removed and the outputs of the three independent runs were pooled using LOGCOMBINER 2.4.7 (Bouckaert et al., 2014). A maximum clade credibility tree was produced from the combined file composed of the three replicate runs using TREEANNOTATOR 2.4.7 (Bouckaert et al., 2014) with mean heights and a posterior probability limit of 0.5 and visualized using FIGTREE 1.4.3 (Rambaut, 2012).

| Biogeographical analysis and ancestral range estimation
The R package 'BIOGEOBEARS' (Matzke, 2013)   c Hoggard et al. (2003). *Plantago macrocarpa has been reported in northeastern Russia (Hultén, 1930), but no vouchers could be located to confirm this.
than the parsimony or Bayesian frameworks originally developed for these models; therefore, the models are called DIVALIKE and BAYAR-EALIKE in BIOGEOBEARS (Matzke, 2013). BIOGEOBEARS also allows for the modelling of founder-event speciation, with the implementation of the + j parameter. We applied seven biogeographical regions in our Flora Iranica (Patzak & Rechinger, 1965). For the cosmopolitan weeds, P. major L. and P. lanceolata, putative centres of origin were used to code their distribution being from Europe and western Asia (based on the floras listed above). Maximum range areas in BIOGEOBEARS was set to two areas, based on there being a maximum of two geographic areas occupied by the extant taxa included in our analyses.
All six possible models (DEC, DEC + j, DIVALIKE, DIVALIKE + j, BAYAREALIKE and BAYAREALIKE + j) were fitted to the data, and the selection of the best-fit model was based on comparing the loglikelihood values and corrected Akaike Information Criterion (Table 3). The results of the ancestral range analysis are visualized as pie charts on each supported node of the tree, signifying the probabilities for all estimated ranges of all possible biogeographic histories.

| Phylogenetic relationships and divergence times for section Plantago
The resulting alignment from concatenating all sequences from the sampled taxa is 4639 base pairs long; however, not all five regions could be amplified for all taxa (see Table 1 Table 1. Maximum likelihood analysis resulted in the same topology, though support was lower at many nodes (Supplementary Information, Figure S3).
Subgenus Plantago is found to be monophyletic (Clade 1; PP = 1.00; Figure 1) and started diverging 8.8 Ma (with an error range of 17.4-2.5 Ma) ( Figure 2). The 14 sampled island taxa were resolved in six different clades (Clades 5, 6, 7, 9, 12, 13) and each is described below. All divergence time estimates presented in the text are followed by error estimates in square parentheses based on the HPD 95% confidence intervals (as shown in Figure 2).
Two taxa classed into section Micropsyllium Decne. form a monophyletic clade (Clade 2; PP = 1.00; 3.9 Ma (7.9-0.8 Ma) which is sister to the rest of the subgenus. The other three taxonomic sections (as defined by Rahn, [1996]) within the subgenus are polyphyletic.
There are up to 12 different clades that can be recognized within subgenus Plantago, all of which have high support. These clades are mainly formed by species that share common geographic distributions rather than taxonomic sections as previously described (Rahn, 1996). Of particular note are the oceanic island species P. aucklandica, P. hedleyi and P. fernandezia, which are taxonomically classed as part of the section Plantago but are phylogenetically clustered with species from other sections in clades that constitute coherent geographic groups.
A clade of European species is found to be the next diverging lineage (Clade 3; PP = 1.00; 6.0 Ma [8.8-1.1 Ma]) and consists of the species P. media L., P. maxima Juss. ex. Jacq. and P. canescens, though the last species also occurs in North America (Table 1). A minor incongruence between the MRBAYES and BEAST trees is the positioning of the European species P. reniformis (compare Figures 2 and 3). In the BEAST analysis ( Figure 2), P. reniformis is placed closest to the European clade, but the relationship between them is uncertain due to low node support. However, in the MRBAYES analyses (Figure 1), P. reniformis is closer to the rest of the subgenus than to the three species in Clade 3, though also with low support (PP = 0.53).
Clade 5 is a highly supported group of Australasian taxa (PP =

| Ancestral range reconstruction
Model selection for the ancestral range reconstruction indicated that the data are best explained by a DEC + j model (LnL = −76.98 and AICc = 160.4; Table 3). The results of the ancestral range reconstruction are depicted in Figure 3, where the probabilities for possible ancestral areas are shown in pie charts at each supported node.
The selection of the DEC + j model indicates that founder-event speciation (i.e. the parameter j) is a major contributor to the currently observed biogeographical patterns within section Plantago.
Interpreting the results of the ancestral range reconstruction at the deeper nodes should be done carefully as incomplete taxon sampling has a major influence on the outcome, and in particular, the presence of phylogenetic uncertainty at some nodes may hinder a correct interpretation. For example, the result presented at the node directly above an unsupported node (indicated with a question mark is older than previously estimated, however within the error margins since earlier works dated the split of Plantago-Littorella from Aragoa at 7.1 or 2.8 Ma Tay et al., 2010a). The dating approach used by Tay et al. (2010a) was similar to the approach used herein (using a Bayesian modelling in BEAST), but only a single calibration point was used (i.e. the young New Amsterdam Island) and therefore recovered a much earlier date of 2.8 Ma (Tay et al., 2010a). Known records for fossil pollen for the genus only extend to the Late Miocene at approximately 6 Ma (Mueller, 1981;Rahn, 1996;Rønsted et al., 2002), though it is possible that the genus is older because fossil ages define only minimum ages. Our divergence analyses were limited to using BEAST, which in comparative dating analyses have previously been shown to give older estimates of dates compared to other methods (Goodall- Considering that less than half of all known species described in subgenus Plantago were included in the current work, further investigation should focus on a more comprehensive sampling of species from the subgenus to resolve these relationships, taxonomy and divergence times with higher certainty.

| Historical biogeography of island endemics
Our sampling in subgenus Plantago was extensive enough to infer a first approximation of the biogeographical history for subgenus Plantago and suggest ancestral ranges of the 14 island taxa we sampled.
The ancestor to all subgenus Plantago likely originated in Europe.
However, the island endemics were found to come from six different lineages each with different ancestral ranges. This suggests that several lineages within subgenus Plantago were successful in dispersing to, and speciating on, oceanic islands, which is not surprising for a group with a global distribution and adaptations to bird dispersal (Birch & Keeley, 2013;de Queiroz, 2005). It confirms earlier theories that the genus Plantago is particularly efficient at dispersing, especially over long ranges Tay et al., 2010a).
Our analyses further show that different patterns of dispersal are inferred for island taxa in Plantago. Geographic proximity is found to be a key factor in determining relatedness and defining the biogeographic histories of some of the island taxa within subgenus Plantago, such that the nearest continents and landmasses were inferred to be source areas for the taxa endemic to Lord Howe Island, Auckland Islands, New Zealand, Tasmania, Juan Fernández Islands and the Falkland Islands. Only in the case of the most remote islands in our study, that is, the Hawaiian Islands and Rapa Iti, and New Amsterdam Island, do we find evidence of extreme long-distance events that defy the rules of proximity, which is in keeping with previous findings on the origins of island floras such as Hawaii (Baldwin & Wagner, 2010).
Interestingly, it is only on these remote islands that multiple Plantago species are known; the remaining islands host single island endemics.
The specific biogeographical findings for each island taxon included in this study are discussed below (and illustrated in Figures 1-3. Plantago fernandezia, from the Juan Fernández Islands off the coast of Chile, and P. moorei from the Falkland Islands were inferred to have been dispersed from ancestors of close geographic proximity in South America. Plantago fernandezia was previously thought to be more closely related to other island taxa from Hawaii and Rapa Iti because it shares more morphological features (such as woody stems typical of insular taxa) in common with them (Pilger, 1937;Rahn, 1996). Our results are, however, in line with the biogeographical histories of other plant lineages from the Juan Fernández Islands having their sources in neighbouring South America (Stuessy et al., 2018). This result also supports the idea that, at least in the case of island taxa, growth forms such as woodiness and other morphological traits (i.e. the presence of only two ovules) can be derived rather than of relict origin and reflect convergent evolution (Carlquist, 1970;Emerson, 2002) rather than phylogenetic relationships.
Plantago hedleyi from Lord Howe Island, P. aucklandica from the Auckland Islands, the mainland New Zealand taxon (P raoulii) and Tasmanian species (P. paradoxa) have their ancestral ranges in Australasia ( Figure 3). Plantago stauntonii from New Amsterdam Island was also found to have ancestral ranges in Australasia, despite the island being situated midway between Africa and Australia, and that  (Baldwin & Wagner, 2010). Despite the seemingly higher possibility of stepping stone dispersal or island hoping driving the dispersal of island taxa in the Pacific, extreme long-distance dispersal events are now considered equally as likely and explain the biogeographies of many globally distributed plant groups (Birch & Keeley, 2013;Dupin et al., 2017;Gillespie et al., 2012;Givnish et al., 2009;Nathan, 2006). Our findings therefore further demonstrate the importance of the Pacific flyway and the occurrence of extreme LDD events for the movement of flora from North America across thousands of kilometres to the Hawaiian Islands and also Rapa Iti (Baldwin & Wagner, 2010). Given that the Hawaiian Islands and Rapa Iti have multiple endemic Plantago species present on them and are among the most extreme with regard to island remoteness in our study group, these island areas may represent the most extreme cases of recent speciation in island Plantago lineages. Further genetic and ecological study of these species may assist in determining what traits are important in not only successful dispersal and colonization to islands but also the subsequent diversification (Baker, 1955(Baker, , 1955Carvajal-Endara, Hendry, Emery, & Davies, 2017).
Although our analyses were sufficient to provide a first approximation of ancestral ranges of island taxa, ancestral range reconstruction at one of the deeper nodes in the tree (shown with a question mark "?" in Figure 3) is highly uncertain due to poorly supported phylogenetic relationships, and increased sampling from subgenus Plantago would be necessary to improve node support and confidence in the ancestral range reconstruction, and clarify the biogeographic histories and relationships between extant Plantago taxa from Africa, Asia and Australasia. For example, our sampling and/or the molecular data produced herein was insufficient to test whether dispersal to Australasia came via the west (Africa) or north (Asia). A more comprehensive sampling of subgenus Plantago would be needed in order to further emphasize that, for biogeographic studies such as this, it is the phylogenetic sampling that is critical, not the taxonomy as proposed by Rahn (1996). Additionally, the definition of biogeographic regions used in our analyses limits the testing of ancestry of insular taxa being in temperate locations (van der Aart & Vulto, 1992). Future analyses could investigate the degree of climate niche matching between island taxa and their ancestral ranges.

| Long-distance dispersal by birds
Our findings support the notion that extant plant species in subgenus Plantago are well adapted to dispersal Tay et al., 2010a), and this was also likely the case for common ancestors. However, the distance that the propagules can disperse may be a factor of the type of birds involved rather than differences in propagule traits.
Due to the closer proximity between some oceanic islands to their near-  (Gillespie et al., 2012;Henshaw, 1910;Jenni & Jenni-Eiermann, 1998;Nogales et al., 2012). Similarly, Plantago may have found its way to New Amsterdam Island (midway between Australia and Africa in the southern Indian Ocean) with the help of larger sea birds or accidental birds capable of flying thousands of kilometres. However, the rarity of LDD events and limitations in studying historical bird movements and behaviour makes it difficult to conclude which birds may have been responsible (Nogales et al., 2012).
Our findings show that, for a globally distributed plant genus that is well suited to LDD by birds, differing scales and modes of dispersal may be equally important in explaining the biogeographical histories. Similarly, differing dispersal modes have also been inferred in explaining the historical biogeographies of other bird-dispersed plant families with unique taxa on multiple oceanic islands (i.e. Asteliaceae [Birch & Keeley, 2013], Rubiaceae [Kainulainen et al., 2017]), and thus, evidence is building that there is not a single LDD model that fits all, but rather that a combination of stepping stone dispersal and extreme LDD can both shape insular floras within closely related plant groups, and that multiple floristic areas can be the sources of closely related island taxa.

| Limitations to diversification
Plantago is an example of a plant group, which is particularly well adapted to long-distance dispersal, and possesses traits such as wind pollination and self-compatibility as well as ability to grow in harsh environments that are conducive to establishment and speciation in insular settings (Baker, 1955). However, the majority of Plantago taxa are single island endemics and thus not as successful in radiation and speciation in the insular setting compared to, for example, ferns that are generally over-represented on islands (Hennequin, Kessler, Lindsay, & Schneider, 2014). As plants with adaptations to LDD such as Plantago are not always over-represented in island floras, it is increasingly being accepted that factors other than dispersal limitations are important for the successful colonization and speciation of insular species (Baker, 1955;Carvajal-Endara et al., 2017;Cheptou, 2012;Heleno & Vargas, 2015). For example, the flora of the Galápagos Islandsof which, 30% are single island endemicswas found to be shaped by habitat filtering rather than by dispersal limitations. Consequently, the match between the species continental climate niche and the island climate was the single best predictor of colonization success (Carvajal-Endara et al., 2017). In the case of orchids, specific traits such as their pollination biology and association with mycorrhizal fungi are considered factors limiting successful colonization (McCormick & Jacquemyn, 2014). The genus Plantago thus remains an interesting case not only to study long-distance dispersal patterns but also to test hypothesis for limitations to successful establishment and radiation in insular habitats.

| CONCLUSION S
Our study provides further insights into the importance of geographic proximity as sources of island flora, even for a group of plants that is arguably well adapted to long-distance dispersal by birds, and that differing scales and modes of dispersal may be equally important in explaining the biogeographical histories of insular species. This further suggests that factors other than dispersal success are important for the establishment and subsequent speciation of insular taxa. This work further emphasizes that classical cladistics approaches to infer closely related species (i.e. using morphological traits) can often mislead the reconstruction of accurate biogeographical histories of island species (de Queiroz, 2005); however, using molecular data to infer ancestral ranges can greatly improve our understanding of biogeographical histories and help elucidate origins, dispersal routes and means in widespread lineages with complex distribution patterns such as Plantago. The genus Plantago is an excellent case to study to further improve our understanding of organismal traits and ecological factors involved in the successful colonization of insular habitats.

DATA AVAILABILITY
All newly generated sequence data for this study have been submitted to GenBank.