A molecular phylogeny of Southeast Asian Cyrtandra (Gesneriaceae) supports an emerging paradigm for Malesian plant biogeography geo‑tectonic and climate processes in historical biogeographical patterns in the region.

The islands of Southeast Asia comprise one of the most geologically and biogeographically complex areas in the world and are a centre of exceptional floristic diversity, harbouring 45,000 species of flowering plants. Cyrtandra , with over 800 species of herbs and shrubs, is the largest genus in the family Gesneriaceae and is one of the most emblematic and species‑rich genera of the Malesian rainforest understorey. The high number of species and tendency to narrow endemism make Cyrtandra an ideal genus for examining biogeographic patterns. We sampled 128 Cyrtandra taxa from key localities across Southeast Asia to evaluate the geo‑temporal patterns and evolutionary dynamics of this clade. One nuclear and four chloroplast regions were used for phylogenetic reconstruction, molecular dating, and ancestral range estimation. Results from the dating analysis suggest that the great diversity of Cyrtandra seen in the Malesian region results from a recent radiation, with most speciation taking place in the last five million years. Borneo was recovered as the most likely ancestral range of the genus, with the current distribution of species resulting from a west to east migration across Malesia that corresponds with island emergence and mountain building. Lastly, our investigation into the biogeographic history of the genus indicates high levels of floristic exchange between the islands on the Sunda shelf and the important role of the Philippines as a stepping stone to Wallacea and New Guinea. These patterns underlie much of the plant diversity in the region and form an emerging paradigm in Southeast Asian plant biogeography.

• The Philippines appear to represent an important secondary source area and stepping stone to Wallacea, Taiwan and Japan, and New Guinea.
• Cyrtandra fits into an emerging paradigm for Southeast Asian plant geography, illustrating the role of geo-tectonic and climate processes in historical biogeographical patterns in the region.

Abstract
The islands of Southeast Asia comprise one of the most geologically and biogeographically complex areas in the world and are a centre of exceptional floristic diversity, harbouring 45,000 species of flowering plants. Cyrtandra, with over 800 species of herbs and shrubs, is the largest genus in the family Gesneriaceae and is one of the most emblematic and species-rich genera of the Malesian rainforest understorey. The high number of species and tendency to narrow endemism make Cyrtandra an ideal genus for examining biogeographic patterns. We sampled 128 Cyrtandra taxa from key localities across Southeast Asia to evaluate the geo-temporal patterns and evolutionary dynamics of this clade. One nuclear and four chloroplast regions were used for phylogenetic reconstruction, molecular dating, and ancestral range estimation. Results from the dating analysis suggest that the great diversity of Cyrtandra seen in the Malesian region results from a recent radiation, with most speciation taking place in the last five million years. Borneo was recovered as the most likely ancestral range of the genus, with the current distribution of species resulting from a west to east migration across Malesia that corresponds with island emergence and mountain building. Lastly, our investigation into the biogeographic history of the genus indicates high levels of floristic exchange between the islands on the Sunda shelf and the important role of the Philippines as a stepping stone to Wallacea and New Guinea. These patterns underlie much of the plant diversity in the region and form an emerging paradigm in Southeast Asian plant biogeography.

Introduction
The islands of Southeast Asia comprise one of the most geologically and biogeographically complex areas in the world (Hall 2002, Lohman et al. 2011 and are the meeting and mixing point of floras and faunas of diverse origins (van Welzen et al. 2011). The area is estimated to harbour approximately 45,000 species of vascular plants (Johns 1995) on more than 20,000 islands (Lohman et al. 2011). It contains the biodiversity hotspots of Sundaland, Wallacea and the Philippines (Brooks et al. 2006), and the mega-diverse island of New Guinea (Mittermeier et al. 2003, Takeuchi 2005, Hoover et al. 2017.
The Malesian region is a complex and intricate mosaic of islands of different origins with a dynamic history over the last 50 million years (Ma) (Hall 2002(Hall , 2012a. In the west of the region are the continental Sunda shelf islands of Sumatra, Java and Borneo, separated by shallow seas. In the centre are the numerous smaller terranes and oceanic islands that comprise the Philippines and Wallacea, and in the east is the Sahul shelf and the large island of New Guinea, which is itself of composite origin. Adding a layer of complexity to this is the changing climate of the past 50 ma (Morley 2012(Morley , 2018, most strikingly the fluctuating glacials and interglacials of the Pleistocene (Woodruff 2010, Morley 2012, which impacted sea levels and the extent of vegetation types (Woodruff 2010, Cannon 2012, Morley 2012. Over the last decade, a number of dated molecular phylogenies (Thomas et al. 2012, Grudinski et al. 2014, Williams et al. 2017) and meta-analyses (van Welzen et al. 2011, De Bruyn et al. 2014, Crayn et al. 2015 have supplemented our knowledge of how current patterns of diversity have been shaped by the geological and climatic history of Malesia. These studies have provided insights into where and when lineages diversified and revealed some remarkable cross-taxon biogeographic patterns (Lohman et al. 2011), including the predominance of west-to-east dispersal patterns (Su and Saunders 2009, Baker and Couvreur 2012, Thomas et al. 2012, Grudinski et al. 2014, Richardson et al. 2014, Crayn et al. 2015 and the identification of Borneo as an ancestral area and centre of diversity. Our current understanding of the biogeographic history of Malesia, however, remains incomplete and there is a need for additional well sampled phylogenies of taxa with distributions across the region (Richardson et al. 2012, Webb andRee 2012).
Cyrtandra is the largest genus in the Gesneriaceae, with over 800 species of herbs and shrubs (Atkins et al. 2013) and is a key component of the herbaceous layer of Malesian rainforest found from sea level to 3000m (Burtt 2001, Atkins et al. 2013. Centres of diversity for the genus in Southeast Asia are Borneo (~200 spp.), the Philippines (~150 spp.), and New Guinea (~120 spp.) (Atkins et al. 2013). Its high species diversity, large number of narrow endemic species, and wide distribution make it an ideal genus for examining biogeographic patterns , Cronk et al. 2005, Clark et al. 2008, Johnson et al. 2017) and investigating the processes which underlie current patterns of biodiversity (Bramley et al. 2004a, Johnson et al. 2015, 2019. The genus is characterised by a combination of two fertile stamens and an indehiscent fruit, which varies from a tough-walled green or brown capsule in the west to a fleshy berry that ripens white or rarely orange in the east of its distribution, particularly in New Guinea and the Pacific (  2019). The flowers are often white but species with pink, red, purple, yellow, green, and orange flowers are also known (Fig. 2).
Earlier phylogenetic studies of Southeast Asian Cyrtandra have focused on particular localities with dense sampling from a small number of locations , Bramley et al. 2004a. Larger scale studies have focused on Cyrtandra diversification across the Pacific (Cronk et al. 2005, Clark et al. 2008, Johnson et al. 2017, and these studies have signalled that the origin of the genus is within the Malesian region. Here, we sample Cyrtandra taxa from key localities across Southeast Asia to construct a well-resolved phylogenetic tree based on one nuclear and four chloroplast regions. We use this to estimate divergence times, ancestral ranges, and dispersal patterns in order to gain insights into the evolutionary history of Cyrtandra. We then consider whether the key patterns recovered for other taxa in Southeast Asia are also seen in Cyrtandra, one of the most species-rich and emblematic genera of the Malesian rainforest understorey.

Taxon sampling
We sampled a total of 192 accessions representing 128 Cyrtandra taxa (Supplementary Table S1), including samples that were representative of all the key regions in Southeast Asia (Fig. 3). Table 1 gives a summary of our sampling against current estimates of species numbers by island for the Southeast Asian region (Atkins et al. 2013). Five taxa from within Didymocarpinae (the same subtribe as Cyrtandra) were selected as outgroups (Aeschynanthus roseoflorus, A. buxifolius, Agalmyla chalmersii, Didymocarpus antirrhinoides, and Loxostigma griffithii). Outgroups were selected to reflect those in recent molecular dating analyses (Johnson et al. 2017, Ranasinghe 2017 to facilitate dating of the nodes using secondary calibration points. We applied names to as many of the samples as possible but many of the samples represent undescribed diversity. There is no single taxonomic treatment for Cyrtandra and therefore species concepts and identifications follow those used in regional treatments (e.g., Cronk, 2001, Bramley et al. 2004b).

Molecular methods
DNA extraction methods and PCR conditions followed Nishii et al. (2019). Details of the primers used for each of the five regions (ITS, matK, trnL-F, psbA-trnH and rpl32-trnL) are given in Table 2. Chloroplast sequence data (matK, trnL-F, psbA-trnH and rpl32-trnL) from two of the outgroup collections, Agalmyla chalmersii and Didymocarpus antirrhinoides, was gifted to the project by Prof Gao Lian Ming, Kunming Institute of Botany, Chinese Academy of Sciences, Yunnan, China.

Phylogenetic analyses
Maximum Parsimony (MP) analyses were initially conducted on individual regions to visually assess were carried out using PAUP v 4.0a163 (Swofford 2002) on unweighted and unordered characters. Alignment gaps were treated as missing data. A heuristic search was carried out using stepwise random addition of 10,000 replicates, with TBR and Multrees on. Statistical

Divergence time estimates
A time-calibrated phylogeny was constructed on the partitioned five-gene dataset using an uncorrelated relaxed lognormal clock in the program BEAST v. 1.10.1 (Drummond et al. 2012, Rambaut et al. 2018. Secondary age calibrations were necessary as there are no unambiguous fossils in the Gesneriaceae family (Wiehler 1983, Clark et al. 2008. Five calibration points were taken from the family-wide phylogeny in Ranasinghe (2017) ( Table 3) and were assigned a lognormal prior following Ho andPhillips (2009) andSchenk (2016) . Five separate runs were carried out, beginning with a random tree and run for 100 million generations under a Yule model of speciation, sampled every 1000 generations. Following Condamine et al. (2015) sensitivity analyses using the Birth-Death tree prior were also run and the results were not significantly changed by choice of tree prior. The results from the

Ancestral range estimation
The R package BioGeoBEARS (BioGeography with Bayesian Evolutionary Analysis in R Scripts) (Matzke 2013(Matzke , 2014) was used to estimate ancestral ranges for Cyrtandra under three historical biogeography methods: DEC (Dispersal-Extinction-Cladogenesis; Ree and Smith, 2008), DIVA (Dispersal-Vicariance Analysis; Ronquist 1997), and BayArea (Bayesian inference of historical biogeography for discrete areas; Landis et al. 2013) models. Descriptions of each of these models and how BioGeoBEARS replicates their key assumptions are given in Matzke (2013). To allow for model comparison, all the models were implemented in a maximum likelihood framework. As Cyrtandra is distributed across a system of islands and shows high levels of narrow endemism, founder event speciation was likely to be a highly relevant biogeographic process (Cowie and Holland 2006, Matzke 2013, Roalson and Roberts 2016, Johnson et al. 2017), so we also explored the influence of founder event speciation in the analysis by including '+J' versions of the models. Due to concerns about the statistical methods in the package, both in terms of how the best fit model is selected and how the +J parameter operates (Ree and Sanmartín 2018), results from all of the six models will be discussed. Given that sampling density can also impact the results, a summary of our sampling rates against current species estimates is presented in Table 1.
The MCC tree was pruned to include only a single representative of each species except for those species that have multi-island distributions and were not monophyletic, such as C. pendula. In these cases, we included one representative from each area (following Johnson et al. 2017Johnson et al. , 2019. For monophyletic species with multi-island distributions, such as C. umbellifera from Taiwan and the Philippines and C. sandei from Java and Sumatra, only one sample was included, and these were coded as present in each respective area in the analysis. Each taxon was assigned a distribution using 13 geographic regions based on a combination of geological information (Hall 2002(Hall , 2012, previous biogeographical studies , van Welzen et al. 2011 and current knowledge of species distributions and relationships in Cyrtandra , Bramley and Cronk 2003, Johnson et al. 2017, Kartonegoro et al. 2018). The 13 regions used were: Hawaii, Marquesas and Society Islands, Japan and Taiwan, Australia, Thailand, Peninsular Malaysia, Sumatra, Borneo, Sulawesi, New Guinea, Solomon Islands, Java and the Lesser Sunda Islands, and the Philippines.
The outgroup taxa were removed so as not to influence the root ancestral area. We set the maximum number of areas to two, as Cyrtandra is characterised by high levels of narrow endemism and only C. pendula is currently recognised as having a range size greater than two areas. The six models (DEC, DIVA-like, BayArea-like, and the '+J' variations of each) were compared for statistical fit using the Akaike Information Criterion (AIC) and a Likelihood Ratio Test (LRT).
Lastly, we used BioGeoBEARS to perform a Biogeographic Stochastic Mapping (BSM) analysis (Matzke 2016, Dupin et al. 2017). This generates simulated histories based on a given biogeographical model, the phylogeny, observed range data, and the calculated ancestral state probabilities at each node averaged over many realisations. The biogeographical events that are possible include within-area speciation, vicariance and dispersal events (range expansions and founder events). Event frequencies were estimated by taking the mean of event counts from 100 BSMs. We ran the BSM exercise on all the models to evaluate the impact that the choice of model has on the event counts.

Phylogenetic relationships
In total, 940 new sequences were generated for this study and our final data matrix contained five gene regions and 5438 aligned base positions. Tree topologies of independent MP analyses of the ITS, trnL-f, psbA-trnH, rpl32, and matK regions were congruent although there was far greater resolution in the ITS dataset than in any of the others individually. The tree based on the concatenated five-gene dataset (Fig. 4,5) largely followed the topology of the ITS tree, with increased support for the relationships between the major clades. There were no incongruences with greater than 80% MP bootstrap support. ML, MP, and BI analyses of the combined dataset resulted in trees with congruent topologies.

Divergence time estimates
In this study, divergence time estimates suggest that the genus Cyrtandra split from its closest relative, Loxostigma, c. 16 Ma (14.18-18.34, 95% HPD) during the early Miocene (Fig. 6). The crown age of Cyrtandra is estimated to be 13 Ma (10.84-15.56, 95% HPD). Summaries of estimated dates are shown in Table 4. Using mean ages, 109 of the 128 taxa included in the analysis split from their most recent ancestor in the last 5 ma (method following Madriñán et al. 2013;Richardson et al. 2014).

Ancestral range estimation
Significant improvement in the likelihood of the three standard models (DEC, DIVA-like, and BayArea-like) was seen when the founder event parameter (+J) was added ( Table 5). Of the six models evaluated, the best fit model was BayArea-like +J. Under all six models analysed, the island of Borneo was recovered as the most likely ancestral area for Cyrtandra (Fig. 7).

Biogeographic stochastic mapping
Under the best fit model, Borneo is resolved as a major source area with ~33% (11.5 events) of all dispersal events originating here (Fig. 8, Table 6). Dispersal from Borneo was most frequent to Sumatra (3.0 events) and the Philippines (2.7 events), followed by Peninsular Malaysia (2.0 events) and Sulawesi (1.5 events). In contrast, Borneo was the recipient of only 0.8% of dispersals from all other areas and no event counts from any single area above 0.1. The Philippines was the second largest source area with 21% of dispersal events originating here. The majority of dispersal events from the Philippines were to the island of Sulawesi (3.9 events), but there were also dispersals from the Philippines north to Taiwan and Japan (1.5 events) and east to New Guinea (0.9 events). The greatest source of dispersals into the Philippines was Borneo (2.7 events). Relatively high levels of dispersal were also recovered between Java, Sumatra, and Peninsular Malaysia (2.5 events from Sumatra to Peninsular Malaysia, 3.0 events from Java to Sumatra, and 1.2 events from Sumatra to Java). In marked contrast to Borneo, Sulawesi emerged as an island of high immigration (receiving 20% of all dispersals) and low emigration, with only 2% of dispersals originating here. New Guinea is resolved as the most likely source of dispersals to the Marquesas (0.9 events), Australia (1.0 event), and the Solomon Islands (1.0 event). The stochastic mapping exercise is more impacted by the choice of model than the ancestral range estimation as the six models rely on different biogeographic processes. In terms of dispersal patterns and potential routes through the region, however, Borneo is always in the top three sources of dispersal in all the models, along with the Philippines and Sumatra. Sulawesi is always interpreted as being a poor source of dispersals under all models. The Philippines is the most important source of dispersals north to Taiwan and Japan under all models. It is the highest source of dispersals south to Sulawesi in all but the two worst fit models where it is the second highest after Borneo; and it is the highest source of dispersals east to New Guinea in all but the two worst fit models where Sumatra, Sulawesi, or Borneo are interpreted as the most important source. New Guinea is the most likely source of dispersals to Australia, the Marquesas/ Society Islands, and the Solomon Islands under all models.

Phylogenetic patterns and relationships
There was no incongruence between the nuclear and chloroplast datasets in our large Malesian sample and hence no detectable evidence of hybridisation. This is in marked contrast to previous phylogenetic studies of Cyrtandra, where extensive hybridisation and/or incomplete lineage sorting has been reported, principally in Hawaii (Pillon et al. 2013, Johnson et al. 2019) but also in the recent radiation of the genus in the Pacific (Johnson et al. 2017). This also differs from the ecologically and biogeographically similar genus Begonia, where the high prevalence of hybrid events is considered to be an important factor driving genomic change and species evolution (Hughes et al. 2018). A recent study of reproductive isolation in four species of Hawaiian Cyrtandra showed that boundaries between sympatric Cyrtandra species are maintained predominantly through postzygotic barriers (Johnson et al. 2015). In Hawaii, all 60 species are the result of a single dispersal event to the archipelago < 5 ma (Cronk et al. 2005, Clark et al. 2008, Johnson et al. 2017, and species are remarkably similar in floral morphology and ecological preference, such that hybridisation is Molecular phylogeny of southeast Asian Cyrtandra    Table 6). The weight of each line indicates the number of predicted dispersal events (both founder and range expansion). All event counts of 0.9 and above are included. Table 6. Summary of all dispersal counts for Cyrtandra averaged across 100 BSMs in BioGeoBEARS. Mapping was performed using parameters from the best-fit model of BayArea-like+J. Colour temperature indicates the frequency of events, with warmer colours indicating more common events (Red = >2.9, Orange = 2.0-2.9, Yellow = 0.9-1.9, Green = < 0.9). The ancestral states (i.e., where the lineage dispersed from) are given in the row, and the descendant states (where the lineage dispersed to) are given in the column. widespread. In contrast, sympatric Malesian Cyrtandra are more likely to be from distantly related lineages (Bramley et al. 2004a), and this is reflected by the higher level of floral divergence (Fig. 2) and apparent lack of hybridisation amongst sympatric species, all of which is congruent with pre-zygotic barriers playing a large role in maintaining species boundaries. The majority of the 36 named taxa that were represented by more than one sample resolved as monophyletic with strong support, including two species with multi-island distributions, C. umbellifera from Taiwan and the northern Philippines (Clade J, Fig. 5) and C. sandei from Java and Sumatra (Clade F, Fig. 5). There were four cases where morphologically similar species pairs resolve as paraphyletic with respect to each other, such as the red-flowered C. clarkei and C. kermesina from Sabah (Clade B, Fig. 4). We might expect to see examples such as this in a recently evolved group where weak gene flow may be a barrier to regaining monophyly following speciation, as has been recorded elsewhere in Gesneriaceae (Hughes et al. 2005). The most striking examples of polyphyly are seen in the widespread and morphologically variable C. pendula and C. picta, distributed across a number of islands. Given patterns elsewhere in the genus, it seems likely that these species names do not represent natural groups but rather a phenetic assemblage which will need to be addressed through further collecting and revisionary work.
Cyrtandra species in New Guinea appear to have much wider distributions than are typically found elsewhere in Malesia . This may indicate a shift in dispersal ability as so many of the New Guinea species have fleshy fruits. Alternatively, this may signal that these taxa represent very recent radiations that have not yet been subjected to range contractions or extinction.

Divergence times
The origin and early diversification of Cyrtandra in the region dates to the mid Miocene (crown age 13 Ma), and most of the current diversity in Southeast Asia is the result of speciation in the last 5 Ma (Fig. 6). These dates are similar to those reported for Cyrtandra in Roalson and Roberts (2016) and Johnson et al. (2017) and significantly younger than those reported in Clark et al. (2008Clark et al. ( , 2009). These last, much earlier, dates were based strictly on the geological ages of islands, an approach which has been shown to be problematic (Renner 2005, Heads 2011). The slightly younger stem age recovered in the present analysis is likely due to the inclusion of Loxostigma, an outgroup taxon that is more closely related to Cyrtandra than those used in earlier Cyrtandra-focused studies.
This pattern of origin in the mid Miocene and young species, mostly the result of speciation in the last 5 My, is reported in other Southeast Asian taxa such as Begonia (Thomas et al. 2012) and Aglaia (Grudinski et al. 2014), and it is highlighted by de Bruyn et al. (2014) in their meta-analysis of regional biodiversity. There are a number of geological factors which are likely to have been drivers of diversification during this period. Approximately 23 Ma the Sunda and Sahul shelves moved closer together, creating land in the centre of the region for the first time (Hall 2002(Hall , 2012a. The subsequent rapid orogenesis on key islands such as Sulawesi and New Guinea (Hall 2002, Hall 2012a also created new habitat. Finally, the climate and sea-level fluctuations of the Pleistocene resulted in cyclic vicariance with frequent habitat fragmentations and amalgamations (Voris 2000, Woodruff 2010, Cannon 2012, Morley 2012. The mean diversification rate for Cyrtandra in Southeast Asia is 0.49 net speciation events per million years, significantly higher than the rate of 0.089 calculated by Magallon and Sanderson (2001) for angiosperms as a whole. The rates of Southeast Asian Cyrtandra diversification are comparable to that of the ecologically similar mega-diverse genus Begonia in the Neotropics (0.5) and in Asia (0.61) (Moonlight et al. 2015). They are slightly slower than in Pacific Cyrtandra, which has a rate of 0.68 (Roalson and Roberts 2016), and significantly slower than lineages of Hawaiian Cyrtandra, in which diversification rates as high as 3.5 are reported (Johnson et al. 2019). In the case of the Pacific radiation, Roalson and Roberts (2016) suggested that geography may have played a significant role in the rapid diversification of taxa, with the emergence of many island archipelagos in the last 5 Ma. Additionally, a transition to fleshy fruits may have aided long-distance dispersal by avian frugivores from source areas in Southeast Asia, followed by diversification in newly colonised island regions.

Ancestral range estimation
The island of Borneo emerges as the most likely ancestral area for the genus and for many of the early diverging clades of Cyrtandra (Fig. 7). Burtt (2001) speculated that Borneo represented the 'original heartland' of the genus based on its high species numbers (c. 200) and the abundant morphological diversity seen here, including richly developed anatomical characters such as sclereids and tracheoids, which 'decrease in all directions' from this centre. The combination of Borneo's large area, relatively stable geological history (Hall 2012, De Bruyn et al. 2014, and extensive areas of rainforest, even during glacial maxima (Cannon 2012), offer a compelling explanation for this lineage accumulation and in-situ diversification. The majority of samples from Borneo in this analysis are from NE Borneo. Increased sampling from elsewhere on the island, particularly the under-collected areas of Kalimantan, will help clarify whether Borneo's important role in the development of biogeographic patterns in the region, is largely due to the key role of NE Borneo's three highest mountains as significant rainforest refugia (as reported for birds; Sheldon 2016) or whether it is more generally attributable to the island as a whole.
With an estimated ancestral range on the Sunda Shelf, Cyrtandra is another example of the increasingly well-documented movement of taxa from the west to east of Southeast Asia, particularly in rainforest lineages (Su and Saunders 2009, Thomas et al. 2012, Grudinski et al. 2014, Crayn et al. 2015. There are examples of dispersal in the opposite direction, such as in Proteaceae (Barker et al. 2007) and Myrtaceae (Sytsma et al. 2004), but there is a distinct asymmetry , Crayn et al. 2015. The west-to-east dispersal appears to have been particularly prevalent from the mid-Miocene onwards as warmer and wetter conditions prevailed, rainforest expanded and extant land emerged east of Wallace's Line , Grudinski et al. 2014, Crayn et al. 2015.
The earliest example of a Sahul−Sunda long distance dispersal in this study is within Clade E at 9.6 Ma (7.64-11.68 Ma, 95% HPD) from Sunda to Sahul. This is comfortably within the time-frame of Sahul−Sunda disjunction events compiled by Crayn et al. (2015). Many of the Cyrtandra taxa east of Wallace's Line are characterised by fleshy berries as opposed to the predominantly dry indehiscent capsules of the Sunda shelf taxa ( Fig. 1; Burtt 2001, Johnson et al. 2017. This change in fruit morphology is possibly associated with a transition in dispersal mode from small mammal dispersal to bird dispersal (Gillett 1967, Burtt 2001. A zoochorous dispersal mode is presented by Crayn et al. (2015) as being the most prevalent for Sunda−Sahul dispersals, with 90% of ancestral species possessing zoochorous propagules, and it seems likely that the transition to fleshier berries has facilitated dispersal of Cyrtandra across the region. As the majority of the species in the Sahul clade (Clade E), including C. vittata and C. bungahijau from New Guinea, C. baileyi from Australia, and C. subulibractea from the Solomon Islands, have fleshy fruits (Gillett 1975, it seems likely that it is the ancestral state for this clade.

Biogeographic patterns Borneo and the islands of the Sunda shelf
Borneo is the source of the highest number of dispersal events for Malesian Cyrtandra, with dispersals to the Philippines, Sulawesi, Sumatra, Peninsular Malaysia, Java, and New Guinea originating here (Fig. 8). This is consistent with De Bruyn et al.'s (2014) characterisation of the island as an evolutionary hotspot defined not only in terms of high species numbers and in situ diversification but also subsequent emigration. High dispersal levels between the islands on the Sunda shelf reflect the shared geological history of these continental islands, which would have formed continuous land during at least some of the glacial maxima (Voris 2000, Hall 2012. At these times there would also have been more extensive areas of rainforest (Cannon et al. 2009, Cannon 2012, facilitating exchange and dispersal in a wet forest genus like Cyrtandra. There are only two samples from the Sunda shelf that fall within the large, predominantly Wallacean, clade (Clade J, Fig 5), and these are both from Central Java, confirming Java's position as anomalous, with links both to the Sunda shelf and Wallacea, as reported by Van Welzen et al. (2011) in their phytogeographic study of the region.

Philippines
The Philippines represent an important secondary source area for Cyrtandra, as well as a stepping-stone for dispersal to some of the more distant regions such as Taiwan and Japan and New Guinea (Fig. 8, Table 6). Dispersal events from the Philippines were recovered south to Sulawesi, north to Taiwan and southern Japan, east to New Guinea, and even south-west to Java. The very different position of the Philippine islands 10 Ma, when Cyrtandra was diversifying, with southern Philippine islands such as Mindanao located much closer to the equator (Hall 2002, offers an explanation for the key role of the Philippines as a route through the region. The Philippines is an area of high species diversity for the genus, with c. 150 species already recorded and new species still being described (Olivar et al. in press). Species from these islands are morphologically very diverse and there are a number of species with fleshy fruits in the Philippines, notably C. hirtigera from Palawan and C. fragilis from Negros and Mindanao, which is congruent with the high vagility of lineages in the Philippine clades. For Cyrtandra, the earliest diverging branch of Philippine taxa are from Palawan and Camiguin (Clade G), dating to just before 10 Ma, suggesting that the Philippines were colonised relatively early in the diversification of the genus and that both short and longer distance dispersal from Borneo played a part in the colonisation of the archipelago. Our results provide some support for the theory that Palawan, or some part of it, could have been above sea level significantly earlier than the start of the Pliocene c. 5 Ma, as proposed for the Palawan Ark Hypothesis (Blackburn et al. 2010, Siler et al. 2012. These results would also require Camiguin to be above sea level earlier than the estimated 2 Ma maximum age for this island (Steppan et al. 2003), although incomplete sampling, particularly from nearby islands in the archipelago, could also explain the discrepancies in dates.

Sulawesi
Sulawesi is an area of significant immigration, being the recipient of 20% of all dispersals, the joint highest in our analysis (Table 6), with dispersals from Java, the Philippines and Borneo. In contrast, dispersal events from Sulawesi to surrounding islands were very infrequent (0.5 events to Sumatra and 0.2 events to the Philippines, Table 6, Fig. 8). It is notable that although fleshy fruited Cyrtandra species, such as the unusual epiphytic C. purpurea, occur on Sulawesi, the majority of species, and all of the most common species, on the island, such as C. hypogaea, C. polyneura and C. kinhoii, are characterised by the drier, tough-walled fruits typical of the Sunda shelf. This low level of emigration and the relative insignificance of Sulawesi in terms of a dispersal route across the region is, however, also seen in other groups such as Begonia (Thomas et al. 2012). Sulawesi is encircled by biogeographic boundaries, including Wallace's and Huxley's line to the west, and Weber's and Lydekker's to the east, suggesting that there are real barriers to dispersal in this area. However, immigration on to the island, particularly from the Philippines and Borneo across the western lines, has clearly occurred more than once in Cyrtandra. Sulawesi's highly dynamic geological history (Hall et al. 2012a, Stelbrink et al. 2012) and the resultant increase in habitat diversity and available niche space for colonising species, offers a possible explanation for the high level of immigration and establishment on the island. Dispersal back across Wallace's line from Sulawesi and successful establishment on the Sunda shelf islands appears to have been difficult for Cyrtandra, possibly due to niche pre-emption in the older, more established forests of those islands. Increased sampling of Cyrtandra from the islands east of Sulawesi, particularly from the Moluccas and New Guinea, will help illuminate whether there is a barrier in this direction or whether this is an artefact of sampling.

New Guinea and the Pacific
New Guinea is resolved as the most likely source of dispersals to Australia, the Solomon Islands, and French Polynesia for Cyrtandra (Fig. 8, Table 6). West-to-east dispersal patterns have been well documented from New Guinea into the eastern Pacific (Keast 1996). Cyrtandra is notable for being the only Malesian Gesneriaceae genus with a distribution that extends significantly into the Pacific (Hilliard and Burtt 2002). The large Malesian genera Aeschynanthus and Agalmyla have wind-dispersed seeds and, despite being highly diverse in New Guinea, do not extend beyond the Solomon Islands and the Louisade Archipelago (Hilliard and Burtt 2002), respectively. Johnson et al. (2017) reported strong asymmetry in the direction of founder events in their study of Cyrtandra in the Pacific, with the majority of dispersals occurring from a west-to-east direction, which they postulated to be the result of bird dispersal. The routes to Australia and the Solomon Islands found here were reported by Johnson et al. (2017) and were suggested by Gillett (1975) based on his extensive knowledge of the Cyrtandra of New Guinea and the Pacific. He reported strong morphological affinities between the Cyrtandra of New Guinea and the Bismarck Archipelago and Solomon Islands, and an attenuation of morphological diversity with increasing distance from New Guinea into the Pacific (Gillett 1975). The longest-distance dispersal event in this study is from New Guinea (Clade I, from Mt Jaya in the far western part of the main cordillera) to the Marquesas/Society Islands. However, this result is most likely an artefact of the relatively low level of sampling from the South Pacific in this study. In the study by Johnson et al. (2017), which included higher density sampling of Pacific taxa, the Marquesan lineage is the result of a single dispersal event from Samoa, while the Society Islands taxa are the result of two dispersal events, one from Fiji and one from Samoa.

Conclusions
Using a robust and well sampled phylogenetic tree to study evolutionary relationships and biogeographical processes in Southeast Asian Cyrtandra, we found evidence in support of (i) Borneo as the most likely ancestral area for the genus, (ii) west-to-east dispersal across the region and into the Pacific, (iii) the Philippines as an important secondary source area and stepping stone to Wallacea, Taiwan and Japan, and New Guinea, and (iv) a mid-Miocene origin for the genus with most of the extant diversity being the result of speciation in the last 5 Ma. These patterns are increasingly well-documented and are beginning to form an emerging paradigm for Southeast Asian plant biogeography. The present study has provided further insight into the fundamental questions of when and where plant diversification took place in Southeast Asia and the role of geo-tectonic and climatic processes in shaping the floristic composition of the area and setting the stage for significant species diversifications. Further work is needed to understand niche evolution and the genomic basis of adaptation to unravel how the massive species richness of Cyrtandra evolved and how it is maintained.
Yunnan, China, for the gift of cpDNA sequences of two of the outgroup samples which was funded by Grant No. 2017-LSFGBOWS-01 from the Large-scale Scientific Facilities of the Chinese Academy of Sciences.
At RBGE, we would like to thank Steve Scott, Sadie Barber, and Nathan Kelso for excellent care of the Cyrtandra living collections, and Michelle Hart, Laura Forrest, and Ruth Hollands for expert advice in the molecular laboratories. Assistance with the programme BioGeoBEARS was generously given by Nick Matzke. Thanks also to Jay Edneil Olivar for enormous help with naming of the Philippine samples included in the analysis.

Supplementary Materials
The following materials are available as part of the online article from https://escholarship.org/uc/fb Table S1. Taxon list for samples in the current study.