Molecular phylogenetics and historical biogeography of the tribe Lilieae (Liliaceae): bi-directional dispersal between biodiversity hotspots in Eurasia

Abstract

Background and Aims

The role played by the Qinghai–Tibet Plateau (QTP) in the organismal diversification and biogeography of plants in the Northern Hemisphere has attracted much attention from evolutionary biologists. Here we use tribe Lilieae (Liliaceae), including primarily temperate and alpine lineages with disjunct distributions in the North Temperate Zone, as a case study to shed light upon these processes.

Methods

Using 191 taxa (five outgroup taxa) comprising more than 60 % of extant Lilieae species across the entire geographical range, we analyse phylogenetic relationships based on three plastid markers (matK, rbcL, rpl16) and nuclear ITS. Divergence time estimation and ancestral range reconstruction were further inferred.

Key Results

The results support a monophyletic Lilieae divided into four clades. Lilium is nested within Fritillaria, which is paraphyletic and partitioned into two clades, New World and Old World, in the chloroplast DNA (cpDNA) analysis. Incongruences between the ITS and cpDNA trees may be explained by divergent ITS paralogues and hybridization. Lilieae originated around 40–49 (28–67) Mya and probably diversified in the QTP region with four major clades that were established during the Oligocene and the Early Miocene. Uplift of the QTP and climatic changes probably drove early diversification of Lilieae in the QTP region. A rapid radiation occurred during the Late Miocene and the Pleistocene, coinciding temporally with recent orogenic process in the QTP region and climatic oscillations. Several lineages dispersed out of the QTP.

Conclusions

Lineage persistence and explosive radiation were important processes for establishing high species diversity of Lilieae in the QTP region. Both long-distance dispersal and migration across Beringia probably contributed to the modern distribution range of Lilieae. Our study shows that biotic interchanges between the QTP region and Irano-Turanian region and the Mediterranean Basin were bi-directional, suggesting the latter was a secondary centre of diversity.

INTRODUCTION

Organismic and environmental processes play a major role in organismal evolution (Richardson et al., 2001; Mao et al., 2012). Since the Cretaceous, one of the most remarkable geological changes in Eurasia has been uplift of the Qinghai–Tibet Plateau (QTP), starting in the early Cenozoic, before the collision of India with Asia (Rowley & Currie, 2006; Royden et al., 2008; Lippert et al., 2014; Renner, 2016; Deng et al., 2017). This orogeny reportedly promoted the diversification of numerous Tibetan organisms (i.e. Gentiana and Rhodiola), exceptional species richness and a high level of endemism harboured in the QTP and adjacent areas (Li and Li, 1993; Wu, 1998; Sun et al., 2012; Zhang et al., 2014; Favre et al., 2015, 2016). Therefore, the QTP region, including the Himalaya–Hengduan Mountains (HHM) region in the south and south-east, has been considered the most important hotspot of biodiversity in the North Hemisphere (NH) (Myers et al., 2000; Marchese, 2015). Elucidating the mechanisms driving plant speciation in this region is of great interest and several spectacular radiations have been reported, involving multiple mechanisms of adaptive radiation (Wen et al., 2014). Local vicariance, secondary contact and ecological speciation events triggered by extensive uplifts of the QTP accompanying climatic fluctuations were proposed as important mechanisms (Liu et al., 2006, 2014; Wen et al., 2014; Mosbrugger et al., 2018). Lineage persistence (the museum theory) as well as explosive radiation theory have also been reported for many plant groups such as Gentiana (Favre et al., 2016), Rhodiola (Zhang et al., 2014) and Saxifraga (Ebersbach et al., 2016). Research investigating the two theories for species-rich plant groups in the QTP region may thus enrich our understanding of the evolutionary origins of biodiversity (Myers et al., 2000). Most biogeographical studies on taxa having their centre of diversity in the QTP region so far have focused on endemic clades or have suffered from biased sampling towards Tibetan taxa, as reviewed by Favre et al. (2015). More case studies need to be implemented to better understand the causes of diversification in the QTP region.

The origins and dispersal patterns of NH temperate plants have been investigated extensively and attracted the attention of numerous botanists for many years (Donoghue and Li, 2001; Donoghue and Smith, 2004). At least three different biogeographical patterns have been revealed. The first pattern suggests that the present disjunctly distributed taxa in the NH resulted from the fragmentation of the once continuous Arcto-Tertiary, Tethyan or boreal floras (Sun et al., 2001; Mao et al., 2010). The second pattern indicates that some genera originated in the QTP region and then dispersed to other NH regions. This pattern appears frequent for temperate plants that are now distributed across Eurasia, and has been associated with the ‘out of QTP’ hypothesis (Jia et al., 2012; Zhang et al., 2014; Favre et al., 2016). The third pattern assumes the origin of the groups in other regions of the world with subsequent widespread diversification after their ancestors reached the QTP (Liu et al., 2002; Tu et al., 2010; Ebersbach et al., 2016). Although the complex biogeographical connections between the QTP and other NH regions have been illustrated up to the present, the areas of distribution of the plants studied were not across all NH areas in most cases (Liu et al., 2002; Tu et al., 2010; Wen et al., 2014; Zhang et al., 2014). The overall role of the QTP for the biogeography and diversification of plants in the NH is still poorly understood (Favre et al., 2015). Intercontinental disjunct distributions between closely related species are a remarkable feature of angiosperm biogeography (Raven and Axelrod, 1974). Extensive molecular phylogenetic and biogeographical studies have revealed disjunct distribution patterns from the NH, especially the Eastern Asia–North America disjunctions (Wen, 1999, 2001; Lu et al., 2011; Zuo et al., 2017). Vicariance associated with global climate cooling and mountain building processes, dispersal across continents via land bridges and long-distance dispersal are the main mechanisms used to explain disjunctive distributions in the NH (Sun et al., 2001; Xiang & Soltis, 2001; Nie et al., 2005; Harris et al., 2013; Wen et al., 2014; Deng et al., 2017). The routes and timing of dispersal and vicariance are found to vary among different lineages, and most previous studies focused on temperate forest elements (Wen, 1999; Nie et al., 2005; Zhang et al., 2014). In this study, we use a widespread group including both temperate and alpine elements, namely tribe Lilieae (Liliaceae), to investigate the biogeographical patterns of plants in the NH and mechanisms of intercontinental disjunct distributions.

Encompassing about 260–300 species in four genera, tribe Lilieae (APG III, 2009; Kim et al., 2013; Kim and Kim, 2013; Peruzzi, 2016) is a member of family Liliaceae, species of which are typically distributed in NH temperate regions. Besides the mountain ranges fringing the QTP, considerable levels of diversity were found in East Asia, Central and West Asia, the Mediterranean Basin and North America. Within Lilieae, the distributions of four genera overlap in the QTP region, which is the main centre of diversity for the tribe, hosting approx. 100 species (Liang, 1995; Liang and Tamura, 2000; Gao et al., 2013). Previous studies focusing on Liliales and Liliaceae have been useful in confirming the systematic position (sister to tribe Tulipeae in Liliaceae) and monophyly of Lilieae (Patterson and Givnish, 2002; Fay et al., 2006; Peruzzi et al., 2009; Givnish et al., 2016). Lilieae therefore is an ideal group to explore the causes of high plant diversity in the QTP region and investigate the biogeographical relationships between the QTP and other NH regions.

In recent molecular phylogenetic studies, tribe Lilieae contains four closely related genera and the infrageneric relationships have been studied extensively (Gao et al., 2013; Day et al., 2014; Yang et al., 2016): Fritillaria, Lilium (including Nomocharis), Cardiocrinum and Notholirion. Fritillaria is the largest genus of the tribe Lilieae and comprises approx. 140 species, occurring in most temperate regions of the NH, and currently divided into eight subgenera (Rix, 2001). These subgenera are mainly supported by molecular phylogenetic analyses (Rønsted et al., 2005; Day et al., 2014). Lilium including Nomocharis is also a large genus consisting of about 120 species and is widely distributed in the NH; seven traditional sections were delimited on the basis of morphological features (Comber, 1949). Phylogenetic studies indicated that most of the species constructed their own clade according to the classification at the section level (Gao et al., 2013). For clarity we will use Nomocharis as a section in Lilium as tentatively suggested by Gao and Gao (2016). Cardiocrinum is endemic to East Asia (including QTP–HHM) and consists of four species ranging across the Sino-Japanese floristic region (SJFR) (Yang et al., 2016). Notholirion contains five species ranging from the western Himalayas to Qinling Mountains and centralized in the QTP region (Liang, 1995; Al-Khayat, 1999; Sun, 2002). The intercontinental disjunct patterns in the tribe Lilieae involve Fritillaria and Lilium. The two genera have a broad distribution in the North Temperate Zone, ranging from North America, Europe, the Mediterranean and Central Asia, to East Asia and northern Asia (WCSP, 2014).

To date, only the evolutionary events in Lilium and Cardiocrinum have been elucidated; Lilium evolved in the QTP and Eastern Asia approx. 14 Mya and Cardiocrinum diversified during the Late Miocene in Central China (Gao et al., 2013; Yang et al., 2016). Patterson and Givnish (2002) considered that Lilieae diversified in the Himalayas roughly at 12 Mya, with only a few species (eight) of Lilieae being included. Givnish et al. (2016) suggested that Lilieae originated in East Asia with divergence beginning at 28 Mya, based on limited taxonomic sampling (22 species) of Lilieae.

Previous phylogenetic examinations of the tribe Lilieae, although highly informative, involved fewer than half of all extant species (Rønsted et al., 2005; Gao et al., 2013; Kim et al., 2013; Day et al., 2014; Givnish et al., 2016). Hence the routes and timings of intercontinental dispersals within this tribe remain obscure. Also, the role of the QTP and surrounding mountain chains in the biogeography and diversification of Lilieae remains unclear. In this study, we constructed phylogenetic relationships among more than 60 % of extant Lilieae species using three chloroplast DNA (cpDNA) markers and nuclear ITS. We aim to obtain a more comprehensive evolutionary framework for Lilieae based on a broad taxon sampling to illuminate the evolution of the tribe. We are also using tribe Lilieae as a case to understand better apparently organismal diversification in the QTP region and the biogeographical relationships between the QTP and other NH regions. Molecular phylogenetic approaches are used to reconstruct the possible historical biogeography of Lilieae and investigate the broad topics mentioned above, as well as to answer the following specific questions: (1) When and where did tribe Lilieae evolve and diversify? (2) What roles did lineage persistence and explosive radiation play in establishing Lilieae’s diversity in the QTP region? (3) What mechanisms might be responsible for this northern temperate intercontinental disjunction of Lilieae?

MATERIALS AND METHODS

Taxon sampling, laboratory work and data handling

Our study represents the most complete sampling of the tribe Lilieae to date, with a total of 199 accessions, representing 191 species (approx. 64–73 % of extant species) in the four genera of the tribe (Supplementary Data Table S1). All eight subgenera within Fritillaria (Rix, 2001) and all seven sections of Lilium (Comber, 1949; Gao et al., 2013, 2016) were included. Also, all four species within Cardiocrinum and four species within Notholirion (excluding Notholirion koeiei) were included. Five species (Amana edulis, Erythronium japonicum, Gagea lutea, Tulipa gesneriana, Lloydia tibetica) in the tribe Tulipeae were designated as outgroups based on previous phylogenetic analyses (Rønsted et al., 2005; Fay et al., 2006; Kim et al., 2013; Petersen et al., 2013; Givnish et al., 2016). For analysis involving cpDNA, 175 accessions were included. Of these 175 accessions, 101 species (109 accessions) were within Fritillaria, 58 species (58) in Lilium and four species (four) in Notholirion and Cardiocrinum, respectively, with 68 accessions newly sequenced. The nuclear ribosomal DNA (nrDNA) data set included 140 accessions, in which 57 species (60 accessions) were in Fritillaria, 72 species (72) in Lilium and four (four) in Notholirion and Cardiocrinum, respectively; of these, 24 accessions were newly sequenced for this study. Multiple accessions were included for Fritillaria species that are widespread (e.g. F. cirrhosa) or whose system location is special (e.g. F. maximowiczii and F. davidii). In total, these species were sampled to cover all of the major distributional areas in Lilieae. For new accessions, 20 Fritillaria species, 43 Lilium species, three Cardiocrinum species, two Notholirion species and three outgroup species were collected between 2006 and 2017 in the field, vouchered and deposited in the Herbarium of Sichuan University (SZ) (see Table S2).

Total DNA was extracted according to the protocols of a plant genomic DNA kit (Tiangen Biotech, Beijing, China). The selected cpDNA regions (matK, rbcL, rpl16) were amplified and sequenced following well-established protocols for Fritillaria (Day et al., 2014); ITS (nrDNA) was sequenced using primers ITS4 and ITS5 (White et al., 1990) and amplified following standard polymerase chain reaction (PCR) protocols (Gao et al., 2013). SeqMan (DNAstar; Burland, 2000) was used to edit DNA sequences and obtain consensus sequences. After alignment in Clustal X version 2.0 (Larkin et al., 2007), manual alignment was conducted.

Phylogenetic analyses

Bayesian inference (BI) and maximum likelihood (ML) analyses were performed for the cpDNA (matK, rbcL and rpl16) and nrDNA (ITS) sequence data separately. Bayesian analyses were performed over 10 million generations with one cold and three incrementally heated Monte Carlo Markov chains (MCMCs) in MrBayes version 3.2 (Ronquist et al., 2012). The best fitting models of nucleotide substitution was determined to be GTR+I+G, based on the Akaike information criterion (AIC) using MrModel Test 2.2 (Nylander, 2004). We sampled one tree per 1000 generations. The first 20 % of generations were discarded as burn-in. A 50 % majority-rule consensus tree was constructed from the remaining trees to estimate posterior probability (PP) values. ML analyses were implemented with RAxML-HPC2 (Stamatakis et al., 2008) on XSEDE at the CIPRES Science Gateway (Miller et al., 2010), with statistical node support estimated via a bootstrap analysis. In addition to these phylogenetic analyses based on the cpDNA and nrITS data, we further constructed ML and BI trees using the same setting for the 116 taxa based on the combined nrITS and cpDNA data (Table S1). We tested for incongruence between the nrITS and cpDNA data sets using the incongruence length difference (ILD) test (Farris et al.,1994) implemented in PAUP* v.4.0b10 (Swofford, 2002).

Divergence time estimation

Divergence times were estimated using a Bayesian method implemented in BEAST v1.8.4 (Drummond et al., 2012). There are currently no well-documented fossils in Liliaceae, and thus fossil constraints were limited to Liliales. Initially we obtained cpDNA sequences (matK, rbcL, rpl16) for 19 Liliales species and added these to our existing 175 Lilieae alignment (data not shown) for divergence time dating. matK and rbcL were aligned in Mega 6.06 and rpl16 was aligned using Muscle (Edgar, 2004; http://www.ebi.ac.uk/Tools/msa/muscle/). The rpl16 alignment process was difficult and showed considerable sequence heterogeneity between Lilieae and other Liliales species (data not shown). Preliminary phylogenetic reconstructions using this dataset produced unlikely results. Thus, we did not use it for further analyses.

Instead, two estimations of divergence times were run. First, we estimated the crown group ages for the tribe Lilieae using sequence data of two cpDNA regions (matK and rbcL) from 35 Lilieae species plus 30 other Liliales species (see Table S3). These selected species represented seven of the ten families in the order. BEAUti (within BEAST) was used to set criteria for the analysis, in which we applied a general time reversible (GTR) nucleotide-substitution model with Gamma+Invariant sites, gamma shape distribution (with four categories) and proportion of invariant sites. An uncorrelated lognormal relaxed clock model and the Yule prior set were used to estimate divergence times and the corresponding credibility intervals. Four calibration points (three from fossils) were used to determine specific node priors. (1) The crown node of Smilax was set to 46 Mya, using a normal prior distribution with standard deviation (SD) 5.3, and 95 % confidence interval (CI) of 37.28–54.72 Mya. This represents a conservative minimal age based on the review by Chacón et al. (2012), given that Smilax-like fossils are known from the Early/Lower Eocene (48.6–55.8 Mya) and the Middle Eocene (37.2–48.6 Mya). In a recent study, Denk et al. (2015) summarized the Cenozoic records of Smilax and indicated Smilax-like fossils might date to the Palaeocene, although unambiguous fossils of smilax date back to the early Middle Eocene. Palaeocene records were difficult to assign to a particular genus with certainty. The results were congruent with the molecular clock analyses of Chacón et al. (2012). (2) Based on fossils of Ripogonum tasmanicum, the stem node of Ripogonaceae was constrained to a minimum age of 51 Mya (Conran et al., 2009), using a lognormal prior distribution with a lognormal mean of zero and a lognormal SD of 1.0. (3) The fossil of Luzuriaga peterbannisteri was used to constrain the crown node of the Drymophila/Luzuriaga clade to a minimum age of 23 Mya (a normal prior distribution, SD 0.5, CI 22–24 Mya) based on the reports in Chacón et al. (2012), which included a more comprehensive sampling of Alstroemeriaceae. They used the leaf fossil described here to constrain the split between Drymophila/Luzuriaga. The dates obtained with the alternative placements of the fossil in the study of Conran and Renner (2014) were congruent and agreed with the hypothesis that the fossil represented an extinct lineage of Luzuriaga that inhabited New Zealand approx. 23 Mya. (4) The root of the tree was constrained to 113 Mya (normal prior distribution, SD 7.3, CI 101–125 Mya) based on the estimate (Givnish et al., 2016) for the crown group of Liliale.

In a second step, we conducted three detailed analyses within Lilieae using a combined cpDNA (matK, rbcL, rpl16), nrITS, and combined nrITS and cpDNA data set, respectively. We used the same settings as in the first step, and the crown ages of the tribe Lilieae obtained from our first analysis were used as ‘secondary calibration’ points, using a normal prior distribution on the root age with mean and 95 % highest posterior density (HPD) to fit those obtained from our first analysis. For BEAST analysis, MCMC runs extended for 50 million generations (first calibration for the crown group ages for the tribe Lilieae) and 100 million (secondary calibration), with parameters sampled every 1000 generations and 10 000 generations, following a burn-in of the initial 10 % of cycles. Three replicate runs were conducted to confirm convergence. Samples were combined, and convergence of chains was checked in Tracer v1.6 to ensure the effective sample sizes (ESS) were all above 200. Maximum clade credibility (MCC) trees were generated in TreeAnnotator v1.8.4 and the MCC chronogram was visualized using the program FigTree v1.4.2.

Ancestral area reconstruction

Seven distributional areas of the tribe Lilieae were delimited according to the regional physical geography and climate history (Tu et al., 2010; Buerki et al., 2012): A, QTP–eastern Himalayas–Hengduan mountains (QTP-HHM); B, Eastern Asia (central, southern, eastern China, Japan, Korea and northern Korea); C, Northern Asia (north-east China, north of Mongolia, Russian Far East and Siberia); D, North America; E, Europe (including northern Europe and European Russia); F, Mediterranean basin (region); and G, Irano-Turanian region (central and western Asia, north-east Africa and north-west China). Areas of distribution of these species were defined according to WCSP (2014) and the distribution of field observations. Widespread ancestral ranges were constrained to a maximum of three distributional areas because extant species of the tribe Lilieae never occur in more than three areas.

The BioGeoBEARS package (Matzke, 2013a, b, 2014; R Core Team, 2017) implemented in RASP 4.0 (Yu et al., 2015) was used to reconstruct the ancestral distributions on the MCC chronogram and trees dataset generated from the BEAST analysis mentioned above (cpDNA, nrITS and combined nrITS and cpDNA data set). We tested the implemented biogeographical models DEC, DIVALIKE and BAYAREALIKE with and without the J-parameter modelling jump dispersal (Matzke, 2013a). A posteriori model testing via AIC was performed to select the best-fit biogeographical model. Outgroups were removed in the ancestral area analyses to reduce limitations posed by the regions of deep groups. We defined three time slices in accordance with global tectonic events with potential relevance to the distribution of the tribe Lilieae, and dispersal multipliers were adjusted for each time slice (see details in Table S4; Favre et al., 2016).

RESULTS

Phylogenetic analysis

The three cpDNA gene regions were combined because no conflicts were observed among partitions. The aligned combined cpDNA data matrix was 3063 bp (matK, 1261 bp; rbcL, 1015 bp; rpl16, 787 bp) long, of which 749 sites (24.5 %) were variable and 443 sites (14.5 %) were parsimony-informative. The total ITS sequence alignment was 662 bp in length and consisted of 443 variable sites (66.9 %) and 332 parsimony-informative characters (50.2 %). The topologies from ML and Bayesian analyses were congruent, and therefore further discussion was based on the Bayesian consensus tree.

The monophyly of the tribe Lilieae was supported strongly by the cpDNA [ML bootstrap support (MLBS) = 100 %; BIPP = 1.00; Fig. 1] and nrITS (MLBS = 100 %; BIPP = 1.00; Fig. 2). In the combined cpDNA analysis, Notholirion (100 %/1.00) and Cardiocrinum (100 %/1.00) were shown to be monophyletic, and the monophyly of Lilium was also supported (97 %/1.00). However, Lilium was nested within Fritillaria. Fritillaria was divided into the New World clade (99 %/1.00) and the Old World clade (98 %/1.00), both of which were monophyletic. The Old World clade and Lilium formed a monophyletic group with moderate support (84 %/0.79), and the group was robustly found to be sister to the New World clade (100 %/1.00). Within the Old World clade of Fritillaria, all six subgenera delimited by Rix (2001) were strongly supported as monophyletic except subgenus Fritillaria, and relationships between subgenera were resolved (Fig. 1). Subgenus Fritillaria was polyphyletic and divided into two clades (Fig. 1). Fritillaria clade A (100 %/1.00) comprised the majority of species and occurred mainly in the Mediterranean Basin and Europe. Fritillaria clade B (91 %/1.00) comprised 18 species sampled mainly from China and formed three monophyletic subclades. The Fritillaria pallidiflora subclade occurred mainly in Xinjiang (north-west China), the F. thunbergii subclade occurred in Eastern Asia and the F. cirrhosa subclade was from QTP-HHM. The F. thunbergii and F. cirrhosa subclades formed a well-supported group (87 %/1.00) that was sister to the F. pallidiflora subclade (91 %/1.00). Within Lilium, the backbone of the phylogeny included three strongly supported lineages forming a polytomy (Fig. 1). One lineage was a disjunctive lineage including European and Asian lilies (97 %/1.00). Another major lineage consisted of taxa endemic to Asia including all sampled QTP-HHM Lilium (88 %/1.00) and the third consisted of North American crown clades (100 %/1.00). Section Archelirion, Pseudolirium and Liriotypus as delimited by Comber (1949) were monophyletic groups with high support. Section Martagon formed the ‘MS’ clade within Sinomartagon II. Section Sinomartagon, Leucolirion and Nomocharis were all polyphyletic.

Fig. 1.

Fifty per cent majority rule consensus tree from Bayesian analysis of combined cpDNA (matK + rbcL + rpl16) dataset. Maximum-likelihood bootstrap support/Bayesian inference posterior probability values are shown near corresponding nodes, while ‘−’ indicates support values of less than 50 %, and ‘*’ represents full support (100 %/1). The generic, subgeneric, sectional and cladistic classification of the tribe Lilieae (Comber, 1949; Rix, 2001; Gao et al., 2013) are indicated to the right of tree and coloured within the tree.

Fig. 2.

Fifty per cent majority rule consensus tree from Bayesian analysis of nrDNA (ITS) data. Maximum-likelihood bootstrap support/Bayesian inference posterior probability values are shown near corresponding nodes, while ‘−’ indicates support values of less than 50 %, and ‘*’ represents full support (100 %/1). The generic, subgeneric, sectional and cladistic classification of the tribe Lilieae (Comber, 1949; Rix, 2001; Gao et al., 2013) are indicated to the right of tree and coloured within the tree.

For ITS analyses, all four genera were supported as monophyletic, but Fritillaria and Lilium had moderate support (81 %/0.91; 54 %/0.99, respectively). Notholirion and Cardiocrinum were sister to each other with weak support (61 %/0.70), Lilium and Fritillaria were also sister to each other with weak support (–/0.70), but otherwise relationships among them were unresolved (Fig. 2). The New World clade of Fritillaria was monophyletic (97 %/1.00) exclusive of F. maximowiczii, and F. maximowiczii was sister to all other Fritillaria species. The Old World clade of Fritillaria was monophyletic, but had weak support (68 %/0.85), yet the relationships among seven subgenera were not well resolved. The ITS data support 11 major crown clades within Lilium (Fig.2), of which three were congruent with sections delimited by Comber (1949): Archelirion (100 %/1.00), Martagon (100 %/1.00) and Liriotypus (99 %/1.00). The combined analysis of cpDNA and ITS data produced a tree similar to that for cpDNA alone, but Fritillaria was strongly supported as monophyletic, being sister to Lilium (Fig. 3). However, the ILD test indicated significant contradiction and incongruence between the nrITS and cpDNA data sets (P = 0.01).

Fig. 3.

Fifty per cent majority rule consensus tree from Bayesian analysis of the combined (nrITS + cpDNA) dataset. Maximum-likelihood bootstrap support/Bayesian inference posterior probability values are shown near corresponding nodes, while ‘−’ indicates support values of less than 50 %, and ‘*’ represents full support (100 %/1). The generic, subgeneric, sectional and cladistic classification of the tribe Lilieae (Comber, 1949; Rix, 2001; Gao et al., 2013) are indicated to the right of tree and coloured within the tree.

Divergence time estimations

Divergence time analyses based on matK+rbcL and four calibration points (three from fossils) resulted in a crown group age of the tribe Lilieae of 28.22 Mya (95 % HPD, 18.90–39.62 Mya). The crown age of Liliaceae was estimated as 73.26 Mya (95 % HPD, 52.72–98.14 Mya) (Fig. S1). Thus, a normal distribution with a standard deviation of 1 and centred about the median 28.2 Mya was applied to tribe Lilieae root node for dating of the cpDNA, nrITS and combined (nrITS + cpDNA) phylogenies. The results obtained from the cpDNA BEAST analysis indicated that Lilieae began to diversify approx. 28.16 Mya in the Middle Oligocene (95 % HPD, 26.23–30.13 Mya) (Fig. 4B and Fig. S2). The stem of Lilieae was dated to 45 (33–61) Mya. The earliest diverging genus, Notholirion, appeared relatively shortly afterwards at 28 (26–30) Mya, Cardiocrinum [25 (21–29) Mya] was the next oldest group, both dating back to the Oligocene. In the Early Miocene, Fritillaria (New World clade) [21 (16–25)], Fritillaria (Old World clade) and Lilium [both 20 (15–24) Mya] evolved. Results of the combined (nrITS + cpDNA) BEAST analysis were generally consistent with that of the cpDNA (Table 1): Fritillaria was monophyletic and evolved in the Early Miocene [21 (17–26) Mya; Fig. S4]. BEAST analysis based on the nrITS phylogeny indicated Lilieae diverged from Tulipeae approx. 40 (28–54) Mya and began to diversify approx. 28 (26–30) Mya (Fig. S3). The earliest diverging genera were Lilium and Fritillaria, both of which appeared in the Late Oligocene [25 (19–29) Mya; weakly supported, 0.74 PP; Fig. S3], and consecutively, Notholirion and Cardiocrinum [both 22 (15–28) Mya] evolved in the Early Miocene (Table 1).

Fig. 4.

Ancestral range estimation and biogeographical scenarios for the tribe Lilieae. (A) Area delineation. (B) Ancestral range estimation based on BAYAREALIKE+J model implemented in BioGeoBEARS. Analysis performed on maximum clade credibility tree from divergence dating analysis (BEAST) using lognormal uncorrelated relaxed clock models for the combined plastid data (matK, rbcL, rpl16). Node support values (posterior probabilities >0.50) are presented. 95 % highest posterior density (HPD) intervals are indicated by grey bars at each node. Cumulative probabilities for estimated ancestral ranges are shown by pies on each node (see Supplementary Data Fig. S5 for a detailed description of the visual representation of results from biogeographical analyses). Colours as in A. Clades are collapsed to aid legibility. (C) Biogeographical scenarios for Lilieae that have experienced dispersal out of the QTP region. Overland movements within continents or via the Bering land bridge (solid arrows) and long-distance dispersal over water or land (hollow arrows). Time scale is shown at bottom. Geological epoch abbreviations: Plio, Pliocene; Quat, Quaternary.

Table 1.

Inferred stem and crown ages (Ma) of major Lilieae clades, and the upper and lower bounds of the 95% higher posterior density (HPD) for those ages based on BEAST analysis. Node support (posterior probability, pp) indicated for each node.

Node	Age estimated in this study
	cpDNA			ITS			cpDNA + ITS
	Mean	95% HPD	pp	Mean	95% HPD	pp	Mean	95% HPD	pp
Lilieae stem	44.49	33.30–60.59	1.00	39.96	28.04–54.26	1.00	49.27	35.29–67.19	1.00
Lilieae crown	28.16	26.23–30.13	1.00	28.13	26.24–30.15	1.00	28.13	26.18–30.07	1.00
Notholirion stem	28.16	26.23–30.13	1.00	21.79	15.05–27.48	0.92	28.13	26.18–30.07	1.00
Notholirion crown	8.88	3.57–16.05	1.00	10.22	5.06–16.59	1.00	10.52	5.44–16.74	1.00
Cardiocrinum stem	24.93	20.47–28.70	0.93	21.79	15.05–27.48	0.92	25.76	21.52–29.15	0.56
Cardiocrinum crown	5.41	1.81–11.00	1.00	13.43	7.44–19.98	1.00	11.51	6.57–17.97	1.00
Lilium stem	19.46	14.94–24.12	0.72	24.51	19.09–28.69	0.74	21.33	16.59–25.85	1.00
Lilium crown	14.92	10.18–19.86	1.00	20.06	14.84–25.62	1.00	16.84	12.29–21.65	1.00
Fritillaria stem	–	–	–	24.51	19.09–28.69	0.74	21.33	16.59–25.85	1.00
Fritillaria crown	–	–	–	19.90	14.12–25.67	0.99	18.12	13.62–22.79	1.00
New World clade stem (Fritillaria)	20.71	15.86–25.03	1.00	17.35	12.35–22.85	0.72	18.12	13.62–22.79	1.00
New World clade crown (Fritillaria)	13.64	8.89–18.85	1.00	7.94	3.98–13.53	1.00	12.53	7.85–17.63	1.00
Old World clade stem (Fritillaria)	19.46	14.94–24.12	0.72	17.35	12.35–22.85	0.72	18.12	13.62–22.79	1.00
Old World clade crown (Fritillaria)	16.78	12.46–21.28	1.00	15.96	11.28–21.19	0.54	15.63	11.53–20.12	1.00
subgenus Davidii stem	16.78	12.46–21.28	1.00	15.96	11.28–21.19	0.54	15.63	11.53–20.12	1.00
subgenus Fritillaria A stem	12.59	9.14–16.60	1.00	11.20	7.49–15.84	1.00	10.11	6.87–13.54	0.99
subgenus Fritillaria A crown	9.33	6.57–12.67	1.00	9.54	6.06–13.53	0.81	7.25	4.78–10.13	1.00
subgenus Rhinopetalum stem	10.64	7.12–14.37	0.98	11.20	7.49–15.84	1.00	10.11	6.87–13.54	0.99
subgenus Rhinopetalum crown	5.53	2.38–9.03	1.00	4.53	1.85–8.42	1.00	4.88	2.25–7.86	1.00
subgenus Japonica stem	10.64	7.12–14.37	0.98	11.49	7.53–16.18	0.93	11.44	7.93–15.11	1.00
subgenus Japonica crown	7.11	4.33–10.66	1.00	–	–	–	–	–	–
subgenus Korolkowia stem	5.37	2.08–9.57	1.00	4.09	1.56–7.24	1.00	4.27	1.90–7.15	1.00
subgenus Theresia stem	8.49	4.00–12.98	0.55	7.89	4.28–11.76	0.71	8.54	5.12–12.24	0.96
subgenus Petilium stem	5.37	2.08–9.57	1.00	4.09	1.56–7.24	1.00	4.27	1.90–7.15	1.00
subgenus Petilium crown	2.52	0.69–5.41	0.98	2.57	0.84–5.08	0.95	2.08	0.76–4.09	1.00
subgenus Fritillaria B stem	11.10	7.27–15.41	1.00	–	–	–	10.08	6.87–13.70	1.00
subgenus Fritillaria B crown	8.32	5.12–11.79	1.00	–	–	–	8.28	5.33–11.46	1.00
section Pseudolirium stem	14.92	10.18–19.86	1.00	14.21	–	0.09	15.57	11.31–19.99	0.82
section Pseudolirium crown	6.37	3.14–10.60	1.00	11.99	7.46–17.19	0.82	7.00	3.71–11.15	1.00
section Archelirion stem	5.69	3.45–8.34	0.89	13.16	7.63–19.40	0.59	10.72	6.34–14.88	0.97
section Archelirion crown	4.89	–	0.20	6.82	3.54–11.33	1.00	3.77	1.28–7.22	1.00
section Leucolirion I stem	6.91	3.82–10.38	1.00	5.88	–	0.48	4.73	2.98–6.80	0.97
section Leucolirion I crown	1.71	0.36–3.75	0.57	4.05	1.80–6.78	1.00	2.67	1.40–4.22	1.00
section Liriotypis stem	8.76	5.22–13.08	1.00	14.45	9.44–20.36	1.00	16.84	12.29–21.65	1.00
section Liriotypis crown	3.37	1.19–6.12	1.00	8.87	4.84–14.01	1.00	6.90	3.44–11.65	1.00
MS clade* stem	6.91	3.82–10.38	0.57	–	–	–	–	–	–
MS clade crown	5.27	2.81–8.50	0.92	–	–	–	–	–	–
section Martagon stem	–	–	–	11.82	8.12–16.48	1.00	7.07	4.77–9.74	1.00
section Martagon crown	–	–	–	6.33	3.44–10.14	1.00	3.85	1.97–6.15	1.00
section Sinomartagon Ⅱ stem	–	–	–	–	–	–	–	–	–
section Sinomartagon Ⅱ crown	–	–	–	–	–	–	–	–	–
section Sinomartagon Ⅲ stem	5.69	3.45–8.34	0.89	10.65	6.97–14.55	0.61	9.32	6.51–12.30	1.00
section Sinomartagon Ⅲ crown	4.24	2.28–6.67	0.59	9.05	5.47–13.28	0.83	7.89	5.15–10.84	0.99
section Nomocharis stem	–	–	–	14.45	9.44–20.36	1.00	12.04	8.78–15.72	0.79
section Nomocharis crown	–	–	–	10.40	6.15–15.31	1.00	8.54	5.59–11.92	1.00
section Nomocharis I stem	6.63	4.03–9.45	0.81	–	–	–	–	–	–
section Nomocharis I crown	4.64	2.27–7.58	0.99	–	–	–	–	–	–
section Nomocharis Ⅱ stem	7.90	5.17–11.34	0.77	–	–	–	–	–	–
section Nomocharis Ⅱ crown	6.18	3.20–9.67	0.63	–	–	–	–	–	–
section Sinomartagon I stem	9.19	–	0.23	14.21	–	0.09	13.61	9.95–17.75	1.00
section Sinomartagon I crown	6.42	3.48–10.10	1.00	10.69	5.96–16.76	0.67	9.40	5.55–13.65	1.00
section Leucolirion Ⅱ stem	10.00	6.46–14.01	1.00	13.16	7.63–19.40	0.59	10.72	6.34–14.88	0.97
section Leucolirion Ⅱ crown	2.89	0.96–5.87	1.00	1.47	0.31–3.60	1.00	1.17	0.39–2.37	1.00

*encompassing section Martagon and a few species of section Sinomartagon (Ⅱ)

Within Fritillaria (Old World clade) and Lilium, the ages of subgenera and sections were between the Miocene and the Pliocene from all three BEAST trees (Table 1), with only minor discrepancies between some nodes. Subgenus Fritillaria was divided into A and B clades. Clade A [stem 13 (9–17) Mya] with subgenus Rhinopetalum and Japonica [sister subgenera splitting at 11 (7–14) Mya] formed a monophyletic clade based on the cpDNA BEAST analysis. The stem of subgenus Fritillaria A inferred from the nrITS and combined (nrITS+cpDNA) analyses was approx. 10–11 (7–16) Mya as was the stem of subgenus Rhinopetalum; the stem of subgenus Japonica including only one species was approx. 11 (8–16) Mya. Fritillaria clade B with subgenus Theresia, Korolkowia and Petilium [sister subgenera splitting at 4–5 (2–10) Mya] formed a monophyletic clade [10–15 (7–19) Mya]. Three remaining main clades include mostly species in eastern Asia and QTP-HHM region, depicting several uncertain phylogenetic relationships and polyphyletic sections, such as Sinomartagon, Leucolirion and Nomocharis. From the cpDNA BEAST analysis, a few species of section Sinomartagon, with all species of section Martagon, formed a monophyletic clade (the ‘MS’ clade) which was sister to section Leucolirion I (weakly supported, 0.57 PP), splitting at 7(4–10) Mya; another species of sections Sinomartagon (Ⅰ, Ⅲ) and Leucolirion (Ⅱ), together with all species of section Nomocharis (Ⅰ, Ⅱ) and section Archelirion, also formed a monophyletic clade [crown age 10 (7–14) Mya]. From the nrITS and combined (cpDNA + nrITS) analyses, species of sections Sinomartagon (Ⅱ, Ⅲ) and Leucolirion (Ⅰ), together with all species of Martagon, formed a monophyletic clade [crown age 11–12 (8–14) Mya]; Nomocharis was a monophyletic clade including four Lilium species [crown age 9–10 (6–15) Mya].

Biogeographical analyses and ancestral area reconstruction

Biogeographical analyses were conducted on the BEAST results based on the chloroplast dataset shown in Fig. 4 and S5. Overall, AIC values showed that the BAYAREALIKE+J model for biogeographical reconstruction yielded the best model fit (Table S5). Inclusion of the ‘jump dispersal’ parameter J significantly improved all models (BAYAREALIKE+J, DEC+J, and DIVALIKE+J) (Table S5), suggesting for Lilieae that the models without founder-event speciation (only accounting for dispersal via anagenetic range expansion) are not adequate to account for all movements to new areas.

From the cpDNA and combined (nrITS + cpDNA) analyses, it could be inferred that the common ancestor of the tribe Lilieae was probably in the QTP-HHM region (Fig. 4, Table S6, Figs S5 and S7), around 26–30 Mya. Early diversification of Lilieae probably took place in the QTP region giving rise to Notholirion (26–30 Mya) and Cardiocrinum (21–29 Mya). Consecutively, the New World clade of Fritillaria (16–25 Mya) originated in the QTP region, and sister groups to the Old World clade of Fritillaria and Lilium (15–24 Mya) both originated in the QTP region based on the cpDNA analysis. However, Fritillaria was monophyletic and sister to Lilium (17–26 Mya), which both also originated in the QTP-HHM based on the combined (nrITS + cpDNA) analysis. From there, Notholirion, Fritillaria and Lilium diversified in situ, giving rise to subgenus Davidii (12–21 Mya) and sections Nomocharis, Sinomartagon and Leucolirion. However, Cardiocrinum colonized an adjacent area, diversifying in a larger region spanning Eastern Asian (B) and the QTP region. Consecutively, several sections in Lilium and subgenera in Fritillaria colonized adjacent areas to the west, the east and the north and led to a large NH distribution range. From the Middle Miocene onward, the area of origin and diversification was unambiguously estimated to be the Irano-Turanian region (G) for subgenera Rhinoperalum, Theresia, Korolkowia, Petilium and Fritillaria B in Fritillaria. Of these subgenera, only Fritillaria B exhibited dispersal back to the QTP region, and then colonized eastern Asia as several species (Figs S5 and S7). Fritillaria A originated in the Irano-Turanian region before spreading to the Mediterranean Basin (F). The Mediterranean Basin is a secondary source area for Fritillaria A (Fig. S5), which diversified there from 10 Mya onward and further dispersed to Europe (E) and back to the Irano-Turanian region (G) in the Pliocene. Subgenus Japonica originated in the Irano-Turanian region but diversified in eastern Asia. From the QTP, northern Asia (C) was colonized by subgenus Liliorhiza (8–19 Mya) which further dispersed to North America (D, 5–15 Mya) (Figs S5 and S7). A few species of sections Leucolirion, Martagon and Sinomartagon in Lilium colonized eastern Asia from the QTP, and the two latter further dispersed to northern Asia (Figs S5 and S7). The stem of section Liriotypis was estimated to have originated in eastern Asia or the QTP-HHM region, although the most probable ancestral areas for the crown group included Europe, the Mediterranean Basin and Irano-Turanian region. Finally, based on our species sampling, North America (D) was colonized by section Psedudolirium directly from the QTP (A, 3–11 Myr).

BAYAREALIKE+J analysis based on the nrITS phylogeny resulted in similar reconstructions but indicated that the QTP-HHM region, Eastern Asia or Northern Asia were the most likely origins of the group (Fig. S6). The earliest divergence event within Lilieae giving rise to Lilium and Fritillaria also occurred in the QTP-HHM, Eastern Asia or Northern Asia. The common ancestor of Notholirion and Cardiocrinum was inferred to be distributed in the QTP-HHM or Eastern Asia, although Notholirion diversified in the QTP-HHM.

DISCUSSION

Phylogenetic relationships within the tribe Lilieae and systematic implications

Our phylogenetic reconstruction of Lilieae corroborates previous phylogenetic findings that Lilieae was a monophyletic group (Rudall et al., 2000; Patterson and Givnish, 2002; Fay et al., 2006; Peruzzi et al., 2009; Givnish et al., 2016). Our cpDNA tree has generated the most comprehensively sampled and well-resolved phylogeny yet of the tribe Lilieae, which represents a good base for further molecular dating and biogeographical analyses. Within Lilieae, three well-supported monophyletic clades were resolved, which were consistent with inferences from previous phylogenetic analyses (Gao et al., 2013, Yang et al., 2016), namely the genera Lilium, Notholirion and Cardiocrimnum. However, Fritillaria was paraphyletic and partitioned into two strongly supported monophyletic clades: New World and Old World. The New World clade including subgenus Liliorhiza was sister to the Old World clade of the Fritillaria–Lilium group; Lilium was nested within Fritillaria with moderate support (Fig. 1). Therefore, our study complements the results of Day et al. (2014) regarding Lilium and two clades of Fritillaria. Relationships within the Old World clade of Fritillaria were generally well resolved, and supported the division of six subgenera proposed by Rix (2001) except subgenus Fritillaria. The cpDNA dataset resolved Lilium into three major lineages forming a backbone polytomy; however, in common with previous studies (Gao et al., 2013), relationships between these were mostly unresolved (Fig. 1).

We detected several phylogenetic incongruences between the ITS and cpDNA data sets (Figs 1 and 2). The relationships among Lilium and the New World and Old World clades of Fritillaria were not supported by the nrITS data (Fig. 2). Additionally, F. maximowiczii was sister to all other Fritillaria species in the nrITS data, whereas it was part of the New World clade in the cpDNA analysis. This incongruence may have resulted from insufficient sampling and resolving power of the ITS data set. Our sequencing of the nrITS region showed divergent ITS paralogues in several species (data not shown), consistent with the results of Day et al. (2014). nrDNA paralogues display polymorphisms in individuals where concerted evolution is incomplete, for example in cases where hybridization is involved (Muir et al., 2001), or where concerted evolution cannot act between paralogues effectively when they are dispersed on non-homologous chromosomes in the genome (Wei and Wang, 2004). These can confound the inference of species relationships in the ITS tree. Except for divergent ITS paralogues, the incongruent phylogenies may mainly be caused by incomplete lineage sorting (ILS) and hybridization/introgression (Degnan and Rosenberg, 2009; Pelser et al., 2010). Because lineage sorting of ancestral genotypes is a stochastic process, it is not expected to follow any geographical pattern (Wu et al., 2015). By contrast, many of the clades in our trees clearly reflect geographical patterns (e.g. Old and New World clades of Fritillaira form reciprocally monophyletic groups in the plastid tree). Therefore, ILS is most probably insufficient to explain the gene-tree incongruence. Morphological and molecular studies have verified that hybridization among Lilium species is an important source of variability (Gao et al., 2013, 2015). So, we suggest that divergent ITS paralogues and hybridization/introgression are the main reasons for the gene tree conflicts of Lilieae. Although a combined analysis of both cpDNA and nrITS data does partly improve this phylogenetic resolution and support Fritillaria was monophyletic (Fig. 3), this result is tentative because the two data sets are not entirely congruent.

Diversification history

From the MCC tree of Liliales, the crown of Liliaceae was estimated to 73 Mya (95 % HPD, 53–98 Mya), which is older than a previously published estimation (67 Mya, 48–92 Mya; Givnish et al., 2016). This incongruence is probably due to our use of three fossil calibrations in Liliales. However, the crown and stem age of the tribe Lilieae from three BEAST analyses were estimated to be 28 Mya (95 % HPD, 26–30 Mya) and 40–49 Mya (95 % HPD, 28–67 Mya), similar to results of Givnish et al. (2016). Hence, during the Middle Eocene, the distinctive features of the tribe arose, but did not diversify until the Middle Oligocene. Ancestral area reconstructions based on the cpDNA and combined (nrITS + cpDNA) phylogeny indicated that Lilieae probably originated in the QTP region (crown Lilieae, Fig. 4; Figs S5 and S7, Table S6). However, BAYAREALIKE+J analysis based on the nrITS phylogeny could not conclusively pinpoint the QTP-HHM, Eastern Asia or Northern Asia as the exact location of most Lilieae lineages during the Oligocene (Fig. S6). The results show an increasing uncertainty towards the root of the nrITS phylogeny, but still revealed an important role for the QTP-HHM. Additionally, the high species diversity and endemism of the four genera of Lilieae in the QTP region make this region a plausible ancestral area of Lilieae. However, the results of Givnish et al. (2016) showed that the crown group of the tribe Lilieae arose in East Asia based on the plastid data. This incongruence is probably due to their incomplete taxon sampling within the tribe Lilieae (only one species per genus). The QTP region has been considered as an active diversification centre for numerous endemic taxa of seed plants (Liu et al., 2006), but also a crucial centre of origin for many other species, such as Gentiana (Favre et al., 2016). Note that our sampling covers more than ~60 % of extant Lilieae based on the cpDNA data. It is possible that including all missing species may modify these results. However, because our sampling is relatively well balanced (per taxon and per area, see Table S7), and there is little difference in the number of missing species in each area, the probability of another area being reconstructed as origin is low.

The palaeobotanical evidence indicates that a period of gradual cooling occurred during the Middle Eocene to Middle Miocene separated by warm intervals (Late Oligocene warming: 27.8–24.5 Mya) (Favre et al., 2015; Hauptvogel et al., 2017). During the Middle Oligocene and Early Miocene, continued uplift of the Himalayas and the QTP alongside global climate cooling and fluctuating may well explain the early diversification of Lilieae into its major extant clades (Favre et al., 2015). Notholirion, Cardiocrimnum, Fritillaira and Lilium all evolved within this time frame in the QTP region based on the cpDNA and combined (nrITS + cpDNA) phylogenies, which showed the QTP region was potentially the centre of diversity of Lilieae species today, suggesting that the QTP acted as a geographical source area for the four genera. Although the nrITS phylogeny analysis could not provide an exact origin for these four genera within Lilieae during the Late Oligocene and Early Miocene, it still revealed an important role for the QTP-HHM (Fig. S6). Orogenesis along the eastern margin of the plateau since the Early Miocene has created high mountains and deep valleys, representing the most diverse and suitable area for Lilieae within the QTP region. In situ radiation of Notholirion [9–11 (4–17) Mya; Figs S5–S7] and Fritillaria evolving to produce subgenus Davidii [16–17 (12–21) Mya; Figs S5 and S7] were probably promoted by the island-like distribution of montane and alpine habitats resulting from surface uplift (Hughes and Eastwood, 2006; Ebersbach et al., 2016). Recently, the interplay of high physiographic heterogeneity and episodes of rapid climatic oscillations (promoting a species pump effect) has been highlighted as a potential driver of alpine diversification (Ebersbach et al., 2017; Mosbrugger et al., 2018). Acting together, diversification in mountains may be enhanced through promoting allopatric speciation and buffering extinction because of topographic complexity (Ebersbach et al., 2016, 2017). This period of in situ radiation of the two clades corresponds roughly to the global climate changes in the Mid-Miocene. Therefore, both high habitat heterogeneity and climatic oscillations were present during the early stages of the radiation of Lilieae in the QTP region. As we do not include all extant Notholirion species in this study, it thus presents minimum estimates. The nrITS phylogeny provides an equivocal origin of subgenus Davidii (Fig. S6), but the endemism of the subgenus in the QTP region makes this region a plausible origin for it. Most species of section Sinomartagon and all of section Nomocharis of Lilium underwent rapid radiations in the QTP region. The former is the largest section in the genus Lilium (Comber, 1949). Our results indicate that Sinomartagon clades Ⅰ and Ⅲ are mainly distributed in the QTP region (with >70 % endemism) and their explosive radiation took place 6–11 (4–17) Mya and 4–9 (2–13) Mya. Nomocharis is a Hengduan Mountains endemic lineage and its rapid radiation occurred 5 (2–8) Mya (Nomocharis Ⅰ) and 6 (3–10) Mya (Nomocharis Ⅱ) from the cpDNA analysis. In the nrITS and combined (nrITS + cpDNA) trees, Nomocharis is monophyletic, including four Lilium species, and its rapid radiation occurred 9–10 (6–15) Mya. The timing of diversification of these lineages coincides with the orogenesis of the QTP and the Hengduan Mountains during the Late Miocene and the Pleistocene (Gao et al., 2013; Favre et al., 2015). Global climatic oscillations have been reconstructed for the Early Pliocene as well as for the Pleistocene (Zachos et al., 2001, 2008; Ebersbach et al., 2017). This corresponds roughly to the period that was suggested for rapid radiations of these lineages. Such physiographic complexity and climatic oscillations provide key opportunities for diversification via allopatric speciation, as well as the evolution of narrow endemics (Hughes & Atchison, 2015; Hughes et al., 2015; Ebersbach et al., 2016, 2017). This inference is similar to conclusions drawn from analyses of other plant groups, such as Saxifraga (Ebersbach et al., 2016), Rhodiola (Zhang et al., 2014) and Rheum (Sun et al., 2012). Another driver that has been suspected to be associated with rapid speciation in Lilium was hybridization mainly for QTP and Hengduan Mountains endemic species, as reported by Gao et al. (2013, 2015), but more data are needed to clarify this. Additionally, the weak phylogenetic relationships within Sinomartagon and Nomocharis were also suggested to be indicative of explosive speciation or rapid radiation in their diversification history (Baldwin and Sanderson, 1998; Yu et al., 2014; Zhang et al., 2014). In summary, the high species diversity of Lilieae in the QTP region was shaped by a combination of local rapid radiation and species persistence in the course of the uplift of the QTP-HHM. Further studies regarding the role of climatic modification and biotic drivers (e.g. key innovation, diversification rates) are needed to complete the picture of species diversification in this region.

Dispersal out of the QTP and intercontinental disjunction within Lilieae

The QTP region constituted a primary centre of origin for Lilieae, and other regional Lilieae species-rich areas elsewhere in the NH were established and affected by several independent dispersal events out of the QTP region (Fig. 4; Figs S5–S7). Thus, Lilieae seems to fit the second hypothesis concerning the biogeographical relationship between the QTP and other NH areas, consistent with the ‘out of QTP’ pattern, which has often been assumed for northern temperate plants (Jia et al., 2012; Zhang et al., 2014; Favre et al., 2016). It is well known that global temperatures began to drop significantly from the Mid-Miocene onwards (Zachos et al., 2001; Zhang et al., 2014). Our dating analysis suggests that the dispersal events took place between the Middle Miocene and the Pleistocene. When temperatures fell, plants adapted to cold habitats may have expanded their range outside of the QTP to other newly available temperate areas in the NH (Zhang et al., 2014; Favre et al., 2016). Two major floristic exchange routes within Lilieae can be recognized. The first route connects Eastern Asia (area B) or onward towards Northern Asia (area C). It mainly occurred within Cardiocrinum, Lilium section Leucolirion, Archelirion and Martagon (see details in Figs S5–S7). This is coincident with biogeographical results for Cardiocrinum (Yang et al., 2016). The second route connects the Irano-Turanian region (area G) to the Mediterranean Basin (area F) and Europe (area E). It mainly occurred within most subgenera of Fritillaria (Old World clade). Extensive uplift of the QTP region together with worldwide cooling since the Middle Miocene (Miao et al., 2012), as well as disappearance of the Tethys Sea (Late Miocene and Early Pliocene) (Lu and Guo, 2014), and prevalence of the East Asian and Indian monsoon climate caused the progressive aridification of Central Asia (Favre et al., 2015). This scenario was also corroborated based on the molecular biogeography of some plants, for example Isodon (Yu et al., 2014), Lagotis (Li et al., 2014) and Cyananthus (Zhou et al., 2013). Fritillaria subgenera Rhinipetalum, Theresia, Korolkowia and Petilium are all distributed in the Irano-Turanian region (Fig. 4; Figs S5–S7), and have a low number of species (subgenera Korolkowia and Theresia are monotypic). The limited species diversification within each subgenus appearing at that time may also be a reflection of aridification in this region. During the Middle Miocene, the closure of the eastern connection of the proto-Mediterranean Sea with the Indian Ocean had significant climatic impacts, causing drier climates over the Iranian plateau and Mediterranean Basin (Rögl, 1999). This aridification trend paved the way for waves of invasion of xerophytic elements into the Mediterranean (Manafzadeh et al., 2014). At the same time, approx. 9 (7–13) Mya, subgenus Fritillaria A (Fig. 4 and Fig. S5) occurred in the Mediterranean Basin. However, the subgenus is the most species-rich, comprising approx. 42 % of Fritillaria taxa (Liang, 1995). During the Late Pliocene, the onset of seasonal climate in the Mediterranean Basin (Thompson, 2005) probably promoted species diversification of subgenus Fritillaria A and several species further dispersed into Europe. The Irano-Turanian region and Mediterranean Basin are the diversity centre and modern abundance centre of Fritillaria (Rønsted et al., 2005). The subgenus Fritillaria samples included from the Mediterranean region in our nrITS and combined (cpDNA+nrITS) phylogenies are very few, and thus estimation of Fritillaria A remains unclear in the two phylogenies. Our study revealed that biotic interchanges between the QTP region and Irano-Turanian region and Mediterranean Basin were not unidirectional (Fig. 4C; Figs S5–S7). The three regions have been considered the most important hotspots of biodiversity in Eurasia (Myers et al., 2000; Marchese, 2015). Therefore, we suggest that the Irano-Turanian region and Mediterranean Basin are secondary centres of diversity of the tribe Lilieae.

Fritillairia subgenus Liliorhiza (New World clade) and Lilium section Pseudolirium are both monophyletic and are mainly distributed in North America (NA). The two lineages illustrate a remarkable disjunction within Lilieae between eastern Asia (EA) and NA. As mentioned previously, there are three explanations usually accepted for EA–NA disjunction: vicariance (fragmentation of the mixed mesophytic forest after the Middle Miocene) (Li et al., 2017), migration across the Bering land bridge (BLB) (Xu et al., 2010; Zhang et al., 2014; Yi et al., 2015), and long-distance dispersal (Tu et al., 2010). For Fritillaria subgenus Liliorhiza, ancestral range reconstruction from the cpDNA and combined (nrITS + cpDNA) phylogenies showed that the North Asian species (F. maximowiczii and F. dagana) formed a grade to F. camtschatcensis ranging from North Asia into North America, and further to the strictly American species (see Figs S5 and S7). Thus, it does not support the vicariance and long-distance dispersal hypotheses. We therefore favour the explanation based on a dispersal scenario across the BLB based on the divergence time of this lineage. The colonization of North America is likely to have occurred 8–10 (5–15) Mya. The BLB was the primary dispersal corridor in the early Cenozoic (66–65 Mya) and was available from 58 to 3.5 Mya to connect northern Asia and North America, thus allowing for biotic interchange between NH landmasses (Brikiatis and Ali, 2014). Northern Asia could have acted as a semi-continuous belt allowing multiple and rapid floristic exchanges between Eurasia and North America (Milne, 2006). Although subgenus Liliorhiza was not monophyletic based on the nrITS phylogeny, the estimated Old World to New World phylogeny is in agreement with two other phylogenies (Fig. S6). The diversification of Lilium section Pseudolirium inferred by our three data is at 6.37–11.99 Mya, which is very similar to the results of Gao et al. (2013), who dated it at 6.85 Mya (the matK phylogeny) and 7.04 Mya (the ITS phylogeny) by using one fossil calibration point. This previous study suggested the Pseudolirium clade colonized North America from the QTP via northern Asia where it underwent local extinction. The age of the Pseudolirium clade estimated in our study supports the hypothesis of a dispersal scenario across the BLB. Dispersal of the tribe Lilieae can occur by both seeds and bulblets. The seeds of Lilieae are flat and narrowly winged (Liang and Tamura, 2000), and thus can be dispersed easily by wind. Bulblet formation sometimes out-competes seed dispersal, especially over long distances (Ronsheim and Bever, 2000). Bulblets may instead be dispersed underground by burrowing animals (Rønsted et al., 2005). Additionally, long-distance dispersal (LDD) probably occurred in two clades within Lilieae (Fig. 4C). Some species of Eurasian lineages within Fritillaria colonized Eastern Asia (Japan) from the Irano-Turanian region where it diversified into subgenus Japonica in the Late Miocene (Figs S5–S7). Another example suggests LDD from Eastern Asia to the Irano-Turanian region onward to Europe (approx. 9–3 Mya) into section Liriotypus within Lilium from the cpDNA tree. However, based on the nrITS and combined (nrITS + cpDNA) trees, Liriotypus shows a second dispersal route connecting the Irano-Turanian region to the Mediterranean Basin and Europe in the Middle and Late Miocene. In general, the wide biogeographical distribution of Lilieae across the northern temperate zone must be interpreted as a result of migration from one contiguous area to another, via existing land bridges and LDD events across land or water barriers.

SUPPLEMENTARY DATA

Supplementary data are available online at https://academic.oup.com/aob and consist of the following. Table S1: List of the tribe Lilieae species included in this study, and corresponding GenBank accession numbers and natural distribution details. Table S2: Herbarium voucher information for the tribe Lilieae species newly sequenced. Table S3: Materials for divergence time estimation of the crown age of Lilieae. Table S4: Priors for biogeographical analyses in BioGeoBEARS, including three time slices and dispersal multipliers matrix per time slice. Table S5: Results from model comparison between six biogeographical models for Lilieae. Table S6: Most likely estimated ancestral ranges for selected Lilieae nodes from three biogeographical analysis under the BAYAREALIKE+J model. Table S7: Summary table of sampling cover of Lilieae. Figure S1: Maximum clade credibility tree of Liliales from BEAST analysis under the Yule tree prior and lognormal uncorrelated relaxed clock models for the combined plastid data (matK+rbcL). Figure S2: Maximum clade credibility tree of the tribe Lilieae from BEAST analysis under the Yule tree prior and lognormal uncorrelated relaxed clock models for the combined plastid data (matK+rbcL+rpl16). Figure S3: Maximum clade credibility tree of the tribe Lilieae from BEAST analysis under the Yule tree prior and lognormal uncorrelated relaxed clock models for the nrITS data. Figure S4: Maximum clade credibility tree of the tribe Lilieae from BEAST analysis under the Yule tree prior and lognormal uncorrelated relaxed clock models for the combined (nrITS + cpDNA) data. Figure S5: Ancestral range estimation in Lilieae. Figure S6: Ancestral range estimation in Lilieae based on nrITS phylogeny. Figure S7: Ancestral range estimation in Lilieae based on nrITS + cpDNA phylogeny.