A transcriptome-based study on the phylogeny and evolution of the taxonomically controversial subfamily Apioideae (Apiaceae)

Abstract

Background and Aims

A long-standing controversy in the subfamily Apioideae concerns relationships among the major lineages, which has prevented a comprehensive study of their fruits and evolutionary history. Here we use single copy genes (SCGs) generated from transcriptome datasets to generate a reliable species tree and explore the evolutionary history of Apioideae.

Methods

In total, 3351 SCGs were generated from 27 transcriptome datasets and one genome, and further used for phylogenetic analysis using coalescent-based methods. Fruit morphology and anatomy were studied in combination with the species tree. Eleven SCGs were screened out for dating analysis with two fossils selected for calibration.

Key Results

A well-supported species tree was generated with a topology [Chamaesieae, (Bupleureae, (Pleurospermeae, (Physospermopsis Clade, (Group C, (Group A, Group B)))))] that differed from previous trees. Daucinae and Torilidinae were not in the tribe Scandiceae and existed as sister groups to the Acronema Clade. Five branches (I–V) of the species tree showed low quartet support but strong local posterior probabilities. Dating analysis suggested that Apioideae originated around 56.64 Mya (95 % highest posterior density interval, 45.18–73.53 Mya).

Conclusions

This study resolves a controversial phylogenetic relationship in Apioideae based on 3351 SCGs and coalescent-based species tree estimation methods. Gene trees that contributed to the species tree may undergoing rapid evolutionary divergence and incomplete lineage sorting. Fruits of Apioideae might have evolved in two directions, anemochorous and hydrochorous, with epizoochorous as a derived mode. Molecular and morphological evidence suggests that Daucinae and Torilidinae should be restored to the tribe level. Our results provide new insights into the morphological evolution of this subfamily, which may contribute to a better understanding of species diversification in Apioideae. Molecular dating analysis suggests that uplift of the Qinghai–Tibetan Plateau (QTP) and climate changes probably drove rapid speciation and diversification of Apioideae in the QTP region.

INTODUCTION

The Apiaceae (Umbelliferae) is a large and readily identifiable family of flowering plants, consisting of c. 430 genera and 3780 species that are treated as four subfamilies: Mackinlayoideae, Azorelloideae, Saniculoideae and Apioideae (Plunkett et al., 2004; Calviño et al., 2016). Although the subfamilies are considered to be monophyletic, many tribes and subtribes traditionally recognized within them are not all monophyletic, especially Apioideae. This subfamily includes c. 404 genera and 2827–2935 recognized species, making it the largest and most taxonomically complex group of Apiaceae (Pimenov and Leonov, 1993). Since the first monograph, Plantarum Umbelliferarum Distributio Nova, of Umbelliferae by Morison (1672), many taxonomic studies of Apiaceae have appeared (Linnaeus, 1753; Koch, 1824; Bentham, 1867; Drude, 1898; Kozo-Poljansky, 1916; Cerceau-Larrival, 1962; Pimenov and Leonov, 1993), and which first identified Apioideae based on fruit characteristics and further divided the subfamily into eight tribes and ten subtribes (Drude, 1898). However, it is becoming more difficult to handle the phylogenetic relationships within Apioideae based on traditional classification systems (Drude, 1898; Heywood, 1986; Shneyer et al.,1992) with more and more species being discovered. With the development of molecular biology, an increasing number of molecular markers have been used for phylogenetic analysis of Apioideae, including: nuclear ribosomal DNA (nrDNA) internal transcribed spacer (ITS) sequences (Downie and Katz-Downie, 1996a; Downie et al., 1998, 2001, 2010; Katz-Downie et al., 1999; Lee et al., 2001; Spalik and Downie, 2001, 2007; Spalik et al., 2004, 2010; Calviño et al., 2006; Ajani et al., 2008; Zhou et al., 2008; Banasiak et al., 2013); chloroplast DNA (cpDNA) genes rbcL (Plunkett et al., 1996a, 1997) and matK (Plunkett et al., 1996b, 1997); intron sequences rpoC1 (Downie et al., 1996b, 1998, 2000), rpl16 (Downie et al., 2000; Zhou et al., 2009) and rps16 (Zhou et al., 2009; Calviño et al., 2016); and cpDNA restriction sites (Plunkett and Downie, 1999; Lee and Downie, 2000). In total, 41 major clades within Apioideae have been identified, of which 21 have been recognized at tribe or subtribe rank, and many morphology-based genera have been split and dispersed among the major clades of the new division based on molecular analysis. However, the relationships inferred from different molecular markers among the major clades were ambiguous and controversial (Zhou et al., 2009; Downie et al., 2010; Spalik et al., 2010; Banasiak et al., 2013; Calviño et al., 2016), and some clades were even revealed as not being monophyletic. Thus, a new classification systemof Apioideae is urgently needed.

The conflict in phylogenetic relationships among major lineages referred in previous studies suggests that hybridization or incomplete lineage sorting (ILS) may be common in Apioideae. Nevertheless, regardless of the molecular data used, the phylogenetic resolution at some nodes is not generally accepted. As a result, there is a need for genes that can clearly resolve the relationships among the major clades at credible resolution. Some recent studies using single copy genes (SCGs) to determine the phylogeny of plant and animal taxa have shown that the resolutions of the phylogenetic trees have improved when compared to previous studies (Dunn et al., 2008; Duarte et al., 2010; Hochbach et al., 2015; Teasdale et al., 2016; Leebens-Mack et al., 2019). SCGs are common across angiosperm genomes and are involved in essential housekeeping functions that are highly conserved across all eukaryotes (De Smet et al., 2013). Shared SCGs in flowering plants are in the unique position of having the closest resemblance to strictly orthologous genes in their genomes, and were considered to contain a greater proportion of phylogenetically informative sites than commonly used protein-coding sequences from the plastid or mitochondrial genomes (Duarte et al., 2010; De Smet et al., 2013). Additionally, SCGs are not subject to concerted evolution as compared with rDNA sequences, and thus allow homologous comparisons (Small et al., 2004; Wu et al., 2006; Li et al., 2008), and their stable copy number and easily assessed orthology have made them suitable for phylogenetic analysis (Yuan et al., 2009). Moreover, SCGs obtained from the transcriptome are coding sequences, and this kind of sequences were suggested to be less subject to ILS than non-coding sequences due to frequent selective sweeps, which tend to remove ILS (Scally et al., 2012). Hence, SCGs may be more suitable than other molecular markers for the phylogenetic analysis of Apioideae.

Deep phylogenetic relationships cannot be reconstructed accurately with small datasets, especially within groups that have experienced rapid evolutionary divergence, ILS and/or reticulate evolution (Dunn et al., 2008; Jian et al., 2008; Han et al., 2014; Zeng et al., 2014). Phylogenetic estimation of species trees based on genomic datasets might resolve branches that were poorly supported based on smaller datasets (Rokas et al., 2003; Dunn et al., 2008; Hochbach et al., 2015; Teasdale et al., 2016; Leebens-Mack et al., 2019). Thus, in this study, we extracted thousands (3351) of shared SCGs from throughout transcriptome datasets and used them to reconstruct the phylogeny of Apioideae. However, it is a challenge to incorporate such large amounts of data into the inference of a species tree, because conflicting genealogical histories often exist in different genes throughout the genome (Degnan and Rosenberg, 2009). Many phylogenetic studies have been based primarily on concatenation-based methods to analyse multiple genes, on the assumption that all genes have the same or similar phylogenies (William and Ballard, 1996; de Queiroz and Gatesy, 2007). However, this revealed that simple concatenation methods could not yield an accurate result for the phylogenetic relationships of an organism (returning incorrect trees with high confidence) in the presence of gene tree heterogeneity (Mossel and Vigoda, 2005; Kubatko and Degnan, 2007; Song et al., 2012), and this method is computationally intensive for large datasets and can be statistically inconsistent under the multi-species coalescent (MSC) model (Roch and Steel, 2014). Numerous methods have been developed to address this shortcoming, and commonly modelled by the MSC model, which has emerged as the standard method for reconstructing species trees in the presence of gene tree discordance due to ILS (Maddison, 1997; Degnan and Rosenberg, 2006; Zhong et al., 2013). Two main categories of model are used. One is Bayesian coalescent-based model, which always co-estimate gene trees and the species tree, such as *BEAST (Heled and Drummond, 2010), BEST (Edwards et al., 2007; Ronquist et al., 2012), BPP (Yang and Rannala, 2014; Flouri T et al., 2018) and revBayes (Hohna et al., 2016). Another group is the most scalable family of MSC-based methods (Sayyari and Mirarab, 2016), which are based on a two-step process in which gene trees are first estimated independently for each gene and then summarized to build the species tree using a summary method. Many such methods are widely used, including MP-EST (Liu et al., 2010), STAR (Liu et al., 2009), NJst (Liu and Yu, 2011) and ASTRAL/ASTRAL-II (Mirarab et al., 2014; Mirarab and Warnow, 2015; Sayyari and Mirarab, 2016). MP-EST can estimate branch lengths in coalescent units; ASTRAL-II estimates species tree topologies, branch lengths and local posterior probabilities; and the remainder only infer the topology.

A reliable species tree often has biological significance and can reflect the evolutionary trend of species. The evolutionary process of speciation includes not only changes to the genetic material, but also changes to the morphological, physiological and behavioural traits of species. Once a robust species tree of Apioideae is available, the true morphological evolution of this subfamily can be explored. Researchers have previously tried to match morphological evolution with the molecular framework in the subfamily (Jiménez-Mejías and Vargas, 2015; Wojewódzka et al., 2019). The Apioideae exhibit considerable morphological diversity, especially in fruit characters, which is a model trait that has received significant attention in studies of the genetic basis of morphological change (Xiang et al., 2017), and they have long played a key role in the classification of Apioideae (Jiménez-Mejías and Vargas, 2015; Zakharova et al., 2016; Liu and Downie, 2017; Lyskov et al., 2017; Wojewódzka et al., 2019). However, the absence of unique characters for most of the major clades and the incongruence between phylogenies make it difficult to describe synapomorphies supporting natural groups. This has led to the interpretation that morphological characters have been highly labile during the differentiation of Apioideae (Jiménez-Mejías and Vargas, 2015). It was also of note that diversity in one character did not imply diversity in another among Apioideae, and that one character and its states may vary tremendously, while another was fixed and totally invariable (Liu et al., 2016). This indicated that diversification in Apioideae has not been stepwise or predictable, but random, and also not gradual but abrupt. Although morphological studies, mostly of fruit morphology, have been carried out on many taxa of Apioideae (mainly focusing on the rank of genera or tribe), there has been no comprehensive study of the whole subfamily.

The aim of the current study was to resolve the long-standing controversy among the major clades of Apioideae and reconstruct a robust phylogenetic relationship based on thousands (3351) of SCGs from transcriptome data of 27 species representing major lineages of Apioideae. This is the first attempt to estimate a species tree of Apioideae through thousands of genes based on the coalescent-based method. We integrate molecular analyses with other data (e.g. fruit morphological traits) to explore the evolution of Apioideae and propose a hypothesis regarding the complex evolution of fruits. Specifically, we attempt to: (1) investigate the phylogenetic relationships among the major lineages of Apioideae using a phylotranscriptomic approach, (2) compare and explain the discordance and agreement between the current species tree topology and those proposed in previous studies, (3) explore the possible evolutionary histories of fruit types in Apioideae, and (4) estimate the divergence time of this group based on SCGs, and roughly infer its possible biogeographical history.

MATERIAL AND METHODS

We have generated a simple analytical process in this study, which is shown in a Flowchart (Fig. 1), and the specific operations are as follows.

Fig. 1.

Flowchart of gene selection and phylogeny construction.

Sampling and data retrieval

Our selected taxa covered from the basal to the distal lineages of Apiaceae subfamily Apioideae (Banasiak et al., 2013), and representative taxa of the 17 major clades were selected based on the framework of Downie et al. (2010). A total of 28 species were sampled including one outgroup, Sanicula orthacantha var. stolonifera, which belongs to Saniculoideae (Table 1), a subfamily that is sister to Apioideae (Chandler and Plunkett, 2004; Plunkett et al., 2004; Calviño et al., 2006, 2016). All the transcriptome data were newly generated except for Daucus carota, for which a whole genome sequence has existed (Iorizzo et al., 2016). The genomic data of D. carota were obtained from Phytozome (http://www.phytozome.net/). Due to difficulties with transcriptome sampling, all the materials were collected from China. All material was taken from young leaves and frozen in liquid nitrogen. Collection details are given in Supplementary Data Table S1. For the transcriptome datasets, total RNA was extracted from young leaves, and library preparation and sequencing were outsourced to Biomarker Technologies (Beijing, China) and Novogene (Beijing, China). Transcriptomes were sequenced on Illumina series device platforms generating 150-bp paired-end reads. We tested the quality of the data and calculated the percentage of GC content, N50 of contigs (bp) and average contig lengths (bp) (Table 1). Clean reads were assembled using Trinity with default settings. Raw reads of the 27 transcriptome datasets have been submitted to the GenBank at NCBI, with accession numbers SRR8863732–SRR8863758.

Table 1.

Information on the 27 transcriptome datasets generated in this study

Taxon	Subfamily	Major clade	Tissue	No. of clean reads	Size of data	Percentage of GC content	Contig N50 (bp)	Average contig length (bp)
Ostericum grosseserratum	Apioideae	Acronema Clade	Leaf	26 428 048	2.115G	39.87	1645	1011.22
Pternopetalum trichomanifolium	Apioideae	Acronema Clade	Leaf	31 501 995	2.555G	39.49	1578	1011.12
Pternopetalum vulgare	Apioideae	Acronema Clade	Leaf	32 721 054	2.715G	39.9	1513	978.1
Foeniculum vulgare	Apioideae	Apieae	Leaf	35 053 593	2.975G	39.27	1866	1158.27
Bupleurum chinense	Apioideae	Bupleureae	Leaf	26 220 538	2.29G	39.78	1181	761.08
Aegopodium podagraria	Apioideae	Careae	Leaf	28 389 906	2.295G	39.35	1487	942.51
Chamaesium paradoxum	Apioideae	Chamaesieae	Leaf	32 327 161	2.62G	39.6	1660	1039.5
Coriandrum sativum	Apioideae	Coriandreae	Leaf	34 531 629	2.895G	39.74	1823	1120.88
Cryptotaenia japonica	Apioideae	Oenantheae	Leaf	30 040 381	2.46G	39.34	2001	1228.02
Oenanthe javanica	Apioideae	Oenantheae	Leaf	30 500 455	2.575G	39.34	1429	940.99
Oenanthe thomsonii	Apioideae	Oenantheae	Leaf	27 399 783	2.295G	39.38	1515	989.51
Haplosphaera phaea	Apioideae	Physospermopsis Clade	Leaf	32 082 500	2.62G	39.31	1835	1174.28
Nothosmyrnium japonicum	Apioideae	Pimpinelleae	Leaf	29 104 885	2.575G	40.32	1290	785.91
Pimpinella diversifolia	Apioideae	Pimpinelleae	Leaf	32 721 054	2.915G	40.16	1440	915.6
Hymenidium davidii	Apioideae	Pleurospermeae	Leaf	25 471 955	2.08G	39.05	1660	1032.29
Cyclospermum leptophyllum	Apioideae	Pyramidoptereae	Leaf	32 146 952	2.675G	39.81	1897	1160.14
Anthriscus sylvestris	Apioideae	Scandiceae (Scandicinae)	Leaf	30 821 940	2.56G	39.81	1760	1090.06
Torilis scabra	Apioideae	Scandiceae (Torilidinae)	Leaf	30 209 020	2.545G	39.61	1609	1036.19
Angelica acutiloba	Apioideae	Selineae	Leaf	26 317 221	2.31G	40.54	1564	951.05
Angelica decursiva	Apioideae	Selineae	Leaf	29 201 429	2.5G	40.77	1243	803.38
Cnidium monnieri	Apioideae	Selineae	Leaf	35 637 090	3.02G	39.4	1690	1017.61
Peucedanum japonicum	Apioideae	Selineae	Leaf	28 900 668	2.34G	39.46	1618	1023.49
Saposhnikovia divaricata	Apioideae	Selineae	Leaf	26 276 265	2.12G	39.66	1690	1074.2
Ligusticum jeholense	Apioideae	Sinodielsia Clade	Leaf	23 800 507	1.91G	39.16	1741	1102.55
Heracleum candicans	Apioideae	Tordylieae (Tordyliinae)	Leaf	30 756 825	2.385G	40.23	1615	987.75
Pastinaca sativa	Apioideae	Tordylieae (Tordyliinae)	Leaf	30 658 501	2.31G	40.12	1789	1115.59
Sanicula orthacantha var. stolonifera	Saniculoideae		Leaf	33 859 594	2.885G	39.45	1778	1116.56

Identification of shared SCGs

To minimize the impact of missing data on the ability to resolve phylogenetic relationships confidently, we included only genes that did not have any taxa missing for further phylogenetic analyses. First, we extracted SCGs of D. carota (Daucinae) from its genome. During this process, the protein data of D. carota were compared against itself using the BLASTN program, with an e-value cutoff of 1e-20. SCGs were then extracted via our in-house script I. We then compared these SCGs against the 27 newly assembled transcriptome datasets using BLAST with an e-value cutoff of 1e-20 and predicted the shared SCGs using OrthoMCL 2.0 (Fischer et al., 2011); the workflow is shown in script II. Additionally, any suggested deleted sequences during implementation of TrimAl v1.4.1(Capella-Gutierrez et al., 2009) were removed from the matrix text file and returned for a second filter. We used the same method to screen out shared SCGs among Group A and Group B respectively.

Phylogenetic reconstruction and species tree estimation

TrimAl was used to adjust all the shared SCGs, and suitable genes were screened out for phylogenetic reconstruction. Within the two species tree estimation methods mentioned above, Bayesian coalescent-based methods readily provide support but are limited to a small number of species and loci, and are computationally intensive (Rannala and Yang, 2017); by contrast, the summary methods are feasible for genome-scale datasets and are statistically consistent. After comparing these two MSC-based methods, we adopted the statistically consistent summary methods as the optimum choice to estimate the species tree of the subfamily Apioideae based on such a large dataset (3351 gene trees). Among the MSC-based summary methods, ASTRAL-II (Mirarab and Warnow, 2015; Sayyari and Mirarab, 2016) has substantial advantages over other coalescent-based methods: it is faster, can analyse much larger datasets (up to 1000 species and 1000 genes) and has substantially better accuracy (Mirarab and Warnow, 2015). In addition, it has been used for many studies based on genome-scale datasets (Cloutier et al., 2019; Leebens-Mack et al., 2019; Allio et al., 2020). Generally, we estimated the species tree using a two-step approach. In the first step, we used RAxML v8 (Stamatakis, 2014) to estimate single-gene trees (rooted and unrooted), where the GTR substitution model and GAMMA-distributed site rates were applied, and support was assessed with 100 bootstrap replicates. In the second step, given the large datasets and the accuracy of species tree estimation (substantial advantages), we generated a species tree based on these single-gene trees under the MSC model preferentially implemented in ASTRAL 5.6.3 (Mirarab and Warnow, 2015; Sayyari and Mirarab, 2016), which estimates species trees from unrooted gene trees, and maximizes the number of quartet trees shared between the gene trees and the species tree. We estimated branch lengths for internal branches (not terminal branches) in coalescent units, which is a direct measure of the amount of discordance in the gene trees, and computed branch support values (which measure support for a quadripartition rather than bipartition). We also computed local posterior probabilities based on gene tree quartet frequencies, which is a reliable measure of accuracy that has very high precision and improved recall compared with multi-locus bootstrapping (Sayyari and Mirarab, 2016). To provide extra per branch information, we used ‘-t 4’ and ‘-t 8’ to annotate the species tree for the main topology and the first and second alternatives, which calculate quartet support (q) and local posterior probabilities (pp) for each branch, respectively.

We also conducted MP-EST and STAR analyses to reduce the chance of missing any true positives (Mao et al., 2019) in the context of ASTRAL-II. MP-EST uses a pseudo-likelihood function of the species tree and estimates branch lengths in coalescent units (Liu et al., 2010), whereas STAR uses average ranks of coalescences (Liu et al., 2009); both of these methods generate bootstrap support values using non-parametric bootstrap techniques. Rooted bootstrapped trees were then uploaded to ‘The Species Tree Analysis Webserver’ STRAW (Shaw et al., 2013) to estimate the species tree using STAR and MP-EST. All three methods are based on summary statistics calculated across all gene trees, with the effect that a small number of genes that significantly deviate from the coalescent model will have relatively little effect on the ability to infer the species tree accurately.

These single-gene trees were further analysed to test the different evolutionary history of the genes. They were clustered into a tree-clade upon neighbour-joining (NJ) analysis. To calculate distances in rooted trees, we considered that the clade near the root is more important than that near the tip of the tree when comparing two rooted trees. Therefore, we developed the normalized sub-node (SN) distance, modified from the normalized RF distance (Robinson and Foulds, 1981) as follows.

For each clade C_k in tree T, let |C_k| be the number of species in C_k, and L(C_k) be the levels (or steps?) from the root of the tree to the root of C_k. $(T_1\stackrel{}{\cap }{T}_2)$ represents the clades shared by the two trees:

$${\displaystyle \begin{array}{ll}{\displaystyle w\left(T_{1,}{T}_2\right)=}& {\displaystyle \sum _{C_k\in T_1}}|C_k|/\mathrm{L}(C_k)\\ & +{\displaystyle \sum _{C_k\in T_2}}|C_k|/\mathrm{L}(C_k)\\ & -2\ast {\displaystyle \sum _{C_k\in (T_1{\displaystyle \stackrel{}{\cap }}{T}_2)}}|C_k|/\mathrm{L}(C_k)\end{array}}$$

The normalized SN distance is derived by dividing w(T₁, T₂) by the maximal possible distance w(T₁, T₂). The normalized SN distance has been implemented in RASP 4.2 (Yu et al., 2020). With the distance matrix, an NJ tree was built using the neighbor program implemented in the PHYLIP package 3.698 (http://evolution.genetics.washington.edu/phylip.html). Topologically similar trees were gathered, forming a branch and suggesting the similar evolutionary history of these corresponding genes. We generated species trees of the different branches of the NJ tree with the methods mentioned above, to compare and explain the discordance and agreement between topologies of these trees. We performed this analysis to explore the impact of different datasets for species tree estimation, and further to extract gene trees that most similar to the species tree, the corresponding genes of which are suitable for further phylogenetic analyses.

Morphological character analysis

A total of 26 fruit samples were collected from the field, except for Ligusticum jeholense (Sinodielsia Clade, provided by Junpei Chen). We were unable to collect fruits of Oenanthe javancia (Oenantheae) and S. orthacantha var. stolonifera (Saniculoideae, outgroup), and thus fruits of the closely related species O. hookeri and S. lamelligera were used. We did not obtain fruits of Aegopodium podagraria (Careae) or Pternopetalum vulgare (Acronema Clade), or fruits of their closely related species. The external characteristics of mature fruits were examined using a stereomicroscope (Nikon SMZ25) and measured using MAML (Altınordu et al., 2016). The fruits were placed on the end of the corresponding branches of the species tree with their real size scaled to the same percentage. For anatomical studies, the fruits were embedded in paraffin and cut at the midpoint, and dissections were stained with safranin–fast green (a few were hand-cut and not stained). Drawings were made based on photographs taken using a Nikon SMZ25 photograph system, and displayed on the corresponding branch of the species tree. We were unable to obtain fruit anatomical of Angelica decursiva and A. acutiloba, thus we provided another typical fruit anatomical of Angelica (A. dahurica) as a complement. To evaluate the evolution of fruit characters, we mapped some onto the final species trees using the maximum parsimony method with Mesquite 3.6 (Maddison and Maddison, 2018). Two key characters were selected, namely the number of developed fruit ribs and the type of commissural primary ribs (marginal ribs), and character scores for each taxon were based on our own observations.

Molecular dating analysis

Previous estimated ages of Apiaceae differ greatly, ranging from 29 Mya to 87.4 (76–99) Mya (Bremer et al., 2004; Bell et al., 2010; Nicolas and Plunkett, 2014; Calviño et al., 2016), which inevitably affected the estimated divergence time of the subfamily Apioideae. The most recent molecular dating analyses of Apioideae were made based on the nrDNA ITS sequence (Banasiak et al., 2013) and cpDNA rps16 intron (Calviño et al., 2016). In the present study, gene trees with similar topologies, broadly consistent with the final species tree, were screened out by script II, and the corresponding genes were extracted. We then concatenated these genes into a large matrix to estimate the divergence time. BEAST v1.8.4 (Drummond et al., 2012) was used to estimate the topology, substitution rates and node ages simultaneously by using a Bayesian Markov chain Monte Carlo (MCMC) approach. The nucleotide substitution model was chosen using Modeltest 3.7 (Posada and Crandall, 1998). We assumed that substitution rates evolved under the uncorrelated lognormal (UCLN) model and the general time reversible (GTR) model for nucleotide substitution. A Yule prior set was used to estimate divergence times and the corresponding credibility intervals. Independent replications each with 10⁸ generations were run with sampling every 10⁴ generations. The stationarity of the chains and convergence of two runs was monitored by Tracer, with the effective sample size of all parameters >200. The chronogram with nodal heights and 95 % highest posterior density intervals (95 % HPDs) was generated with TREEANNOTATOR, with the first 1000 trees being discarded as burn-in. Two calibration points from fossils were used to determine species node priors – we mainly used the pollen fossil adopted by Banasiak et al. (2013). The first calibration point was placed at the stem node of Bupleureae (Gruas-Cavagnetto and Cerceau-Larrival, 1984) and was constrained to a lognormal distribution with a lower bound (offset) of 33.90 Mya (the end of the Priabonian). We also extended the strategy used by Banasiak et al. (2013)—adjusted the mean to set the upper bound of the distribution (95 % quantile) to 58.7 Mya (corresponding to the beginning of the Thanetian) instead of choosing arbitrary parameters. The second point with a similar constraint was imposed for the stem node of Pleurospermeae (Gruas-Cavagnetto and Cerceau-Larrival, 1984), and the upper bound was constrained to the beginning of the Ypresian (55.8 Mya). We abandoned the third fossil given by Banasiak et al. (2013), which was suggested to be attributed to a monophyletic group comprising Scandiceae, Smyrnieae, Aciphylleae and the Acronema clade, because (1) information regarding this fossil has not been published already; (2) the study by Calviño et al. (2016) showed that Scandiceae, Smyrnieae and Aciphylleae could not be used as a monophyletic group, which made it is difficult to choose a reasonable position for calibration. This suggests that information regarding this fossil is not reliable, especially given the small number of species we used. Arbitrary values of 0.5 for the standard deviation were chosen; all information relating to the two calibration points is given in Table 2.

Table 2.

Temporal constraints used to estimate divergence times for Apioideae (Mya); the standard deviation of the lognormal distribution was set to 0.5 for all calibration points

Stem node (fossil reference)	Offset	Mean	Median	95% quantile
Bupleureae (Gruas-Cavagnetto and Cerceau-Larrival, 1984)	33.90	2.389	44.80	58.71
Pleurospermeae (Gruas-Cavagnetto and Cerceau-Larrival, 1984)	33.90	2.264	43.52	55.80

RESULTS

Sequencing of transcriptome datasets and identification of shared SCGs

We sequenced 27 new transcriptome datasets covering most of the lineages of Apioideae, especially lineages lacking sequenced genomes. The lineages we selected were distributed uniformly in the phylogenetic trees of Apioideae reconstructed by Banasiak et al. (2013) based on ITS, eight of which belong to the apioid superclade (Downie et al., 2010). The transcriptome datasets were large enough and contained sufficient information for subsequent analysis, and the quality of these datasets were good as revealed by percentage of GC content, the N50 of contigs and average contig lengths (Table 1). The percentage of GC content varies slightly (ranging from 39.05 to 40.77 %), and the average length of contigs varied from 761.08 to 1228.02 bp, along with N50 values ranging from 1181 to 2001 bp. We performed a series of analyses on the transcriptome datasets and genomic data. A total of 8886 SCGs were screened out from the D. carota genome initially, 3369 of which were identified as shared SCGs through BLAST and MCL analyses. Within these shared SCGs, 18 were filtered out through TrimAl processing as being unsuitable for phylogenetic tree construction. As a result, we finally identified 3351 shared SCGs for the phylogenetic reconstruction of Apioideae. Subsequently, we screened out 11 shared SCGs for divergence time estimation, and the final alignments of these 11 concatenated SCGs reached 19 855 bp.

Phylogenetic relationships among major lineages of Apioideae

A total of 3351 rooted/unrooted gene trees were obtained through RAxML, bootstrapped trees were analysed under all three methods and an extra analysis was implemented in ASTRAL-II with the best maximum-likelihood (ML) trees. The same topology was retrieved regardless of the bifurcate tree-building method used (including MP-EST, STAR and ASTRAL-II) (Fig. 2). All tree-building methods lent strong branch support values for every branch of the species trees, except branch IV, which shows high branch support values in STAR (100) and ASTRAL-II (97, 1.0) but a moderate MP-EST branch support value (67). Quartet support analysis in ASTRAL-II suggested that five branches (I–V) have support values <50% for the main topology (Fig. 2; Supplementary Data Fig. S1), indicating a considerable gene tree conflict around these branches (Sayyari and Mirarab, 2016). However, the extremely high local posterior probabilities (pp1 = 1, pp2 = 0, pp3 = 0) for each branch suggest a very high precision of our species tree.

Fig. 2.

Species tree generated using ASTRAL-II based on 3351 single copy gene trees. Numbers or asterisks above branches are branch support values for MP-EST, STAR and ASTRAL-II with bootstrapping analyses and local posterior probabilities for ASTRAL-II with best maximum-likelihood trees, respectively, with asterisks denoting maximum support in all four analyses. ASTRAL-II measures branch lengths in coalescent units (the scale bar shown corresponds to two coalescent units) for internal branches and not terminal branches (branch lengths of terminal branches are therefore arbitrary and meaningless). The pie charts show respective quartet support for the main topology, and the first and the second alternative topology.

Broadly, our species tree indicates that all of the apioid superclade groups are strongly supported, forming Group A. Particular branches in this group with almost the same quartet support for three topologies (the main topology, the first alternative,and the second alternative) are III and IV, associated with the Sinodielsia Clade (L. jeholense) and Apieae (Foeniculum vulgare), with the following quartet support values: III (q1 = 0.37, q2 = 0.32, q3 = 0.31) and IV (q1 = 0.37, q2 = 0.29, q3 = 0.34) (Fig. 2; Supplementary Data Fig. S1). Additionally, branch lengths for these two branches are extremely short. Branch V, which links the internal branches in Selineae, is supported by only 48% of the gene trees in the main topology.

To explore the relationships among major lineages of Group A, we performed gene resampling analysis. Two taxa were added in this study, Angelica dahurica and A. laxifoliata, and 3793 shared SCGs were obtained for the phylogenetic reconstruction of Group A. The species tree of Group A was estimated using ASTRAL-II based on both ‘bootstrapped trees’ and ‘best ML trees’. The topology of the species tree of Group A (Supplementary Data Fig. S2) was consistent with this group within the Apioideae species tree estimated by the former (Fig. 2). However, when compared with the species tree of Apioideae, the species tree of Group A shows significant decrease in branch support for branch III, with only moderate (62) and low support (0.47), and the branch length is significantly shortened (Fig. S2). Quartet support analysis suggested that branches III–V had lower gene tree support than other branches (Figs S2 and S3).

Furthermore, Group A and Group B are strongly supported forming a sister clade, with Group B comprising Daucinae (D. carota), Torilidinae (Torilis scabra), Scandicinae (Anthriscus sylvestris), and the Acronema Clade (including P. trichomanifolium, P. vulgare and Ostericum grosseserratum) (Fig. 2). Daucinae and Torilidinae are sisters with moderate quartet support for the main topology (q1 = 0.64), and subsequently forming successive sister clades with the Acronema Clade (q1 = 0.59) and Scandicinae (q1 = 0.77) (Supplementary Data Fig. S1). Note that Daucinae, Torilidinae and Scandicinae were previously assigned to the tribe Scandiceae (Downie et al., 2010; Banasiak et al. 2013), but herein these three clades do not constitute a monophyletic group. We therefore resampled 4257 genes for Group B analysis separately, and this again revealed that the three subtribes could not form a monophyletic group (Figs S4 and S5), and the relationships among these four major clades within Group B were without any changes. Group C is composed of a natural monophyletic tribe Oenantheae, containing O. javanica, O. thomsonii and Cryptotaenia japonica, and the internal branches have very high quartet support values (Fig. S1). In addition, our species tree strongly supports Chamaesieae (Chamaesium paradoxum) as the most basal lineage of the Apioideae we studied here, not the Bupleureae (Bupleurum chinense). Our species tree shows that Groups A, B Group C exist as successive sister groups clustering a stabilized distal branch D [Group C, (Group A, Group B)], and this branch D along with the Physospermopsis Clade (Haplosphaera phaea), Pleurospermeae (Hymenidium davidii), Bupleureae and Chamaesieae then form successive sister clades.

In further analyses, we clustered all the single-gene trees according to distance into an NJ tree (Fig. 3; Supplementary Data Fig. S6), which generated four main branches, Rt, Pt, Bt and Ot, for which the number of gene trees was 405, 502, 1184 and 1260, respectively. The species trees of these four datasets showed broadly similar topologies except for the basal clades (Fig. 3; Figs S7–S16). The difference between species trees of the two richer gene trees branches, Bt (Fig. S7) and Ot (Fig. S10), lies in whether Bupleureae or Chamaesieae is the most basal clade. The Bt species tree supports Chamaesieae at the most basal clade, which is consistent with the final species tree (i.e. the species tree estimated based on 3351 gene trees), whereas the Ot species tree has Bupleureae as the most basal lineage. Within the Rt species tree (Fig. S12), Chamaesieae is still the most basal lineage of the subfamily Apioideae, and Pleurospermeae then separated earlier than Bupleureae. The Pt species tree (Fig. S14) shares the same topology with the Bt and final species trees but with lower quartet support values for each branch, especially for the most basal internal branch, and this branch also has depressed branch support (93/0.59). We used the ‘-q’ option in ASTRAL-II to score the species trees of the four branches and the final species tree, with the normalized quartet score used as a measure of discordance for the gene trees: the higher number, the less discordant are the gene trees (Sayyari and Mirarab, 2016). In our analyses, the normalized quartet score of the Bt tree is the highest (0.859) and Pt is the lowest (0.650), while those of Ot (0.832) and Rt (0.818) are slightly lower than Bt (Table S2). This indicates that the Pt datasets has the highest ILS level, whereas Bt has the lowest.

Fig. 3.

A cluster of all the 3351 gene trees and the species trees of the four branches. (A) The neighbor-joining tree of gene trees; all trees gathered into four branches marked by Rt, Pt, Bt and Ot. (B) The species trees of these four branches. Numbers in the first column above branches are quartet support values, and the asterisks and numbers in parentheses above branches are branch support values for ASTRAL-II with bootstrapping analyses and local posterior probabilities for ASTRAL-II with best maximum-likelihood trees, respectively, with asterisks denoting maximum support in both analyses.

Fruit characters of Apioideae

The sampled taxa represent a wide array of the fruit morphologies found in the subfamily. Photographs of fruits and sections through them are shown in Fig. 4. Fruit character explanations are given in Fig. 4A, for which we base all of the characters described below. The Apioideae display a wide spectrum of fruit morphotype, varying greatly in fruit size, number and type of ribs, shapes of fruit transverse sections, etc. (Fig. 4B). Superficial observations of fruit size suggested that fruits of the basal lineages were general smaller than those at the tip of the species tree, including Chamaesieae, Bupleureae, Oenantheae and some taxa of Group A, with these taxa existing as small herbs; larger fruits appear within tall, robust Umbellifers, such as Selineae, Tordylieae, the Acronema Clade (Ostericum), and some basal lineages such as Pleurospermeae and the Physospermopsis Clade. Anatomical studies indicated that these fruits may be roughly grouped into three categories according to rib number and type: fruits with co-developed primary and secondary ribs, having a total of nine ribs (Chamaesieae); fruits with unique prominent secondary ribs (Daucinae and Torilidinae), generally having four ribs; and fruits of the remaining groups with developed primary ribs and degradative secondary ribs, these always containing five ribs (Fig. 4; Supplementary Data Fig. S17). The type of marginal rib within the third fruit category varies widely from filiform to winged, including filiform, subtriangular, narrowly winged and widely winged (Fig. S17). Fruits with widely winged marginal ribs are particularly common in tribes Selineae, Tordylieae, and some members of the Acronema Clade (Ostericum), and dorsal and lateral primary ribs of these fruits are always filiform or slightly winged (Fig. 4). Although fruits of Pleurospermeae (Hymenidium) and the Physospermopsis Clade (Haplosphaera) also have winged marginal ribs, the marginal ribs are only slightly developed compared with the other primary ribs. This kind of fruit also appears in the basal lineages of Group A (Apieae and Sinodielsia Clade). Fruits with subtriangular marginal ribs occur in Group C, and have seeds that are always covered by corky mesocarp and waxy pericarp. Fruits are dramatically diverse in Group B, including fruits with filiform marginal ribs (some taxa of the Acronema Clade, e.g. Pternopetalum), with widely winged marginal ribs as mentioned above (Ostericum), without any prominent ribs (Scandicinae, e.g. Anthriscus) and with developed secondary ribs (Daucinae and Torilidinae). Specialization of fruit ribs is common in Apioideae, mainly falling into two types: spiny appendages (sometimes winged) that occur on secondary ribs, found only in Daucinae and Torilidinae; and specialization of the primary ribs, especially the marginal ribs, which generally extend into wings. This specialization occurs in most fruits of the subfamily Apioideae, especially in Group A and Group B. Fruits of the outgroup, S. lamelligera (Saniculoideae), without any prominent ribs, are densely covered with irregularly arranged straight bristles.

Fig. 4.

Morphological variation and evolutionary histories of Apioideae. (A) Descriptions of the fruit characters considered in this study. (B) Morphologies of fruit appearance displayed at the end of the corresponding branches of the bifurcated tree, in the context of the phylogeny shown in Fig. 2, with all fruits scaled to their true size at the same percentage. Taxa in the purple, red and green boxes form Group A, Group B and Group C, respectively. D refers to distal branch D. A1: Angelica decursiva; A2: A. acutiloba; A3: Peucedanum japonicum; A4: Saposhnikovia divaricata; A5: Cnidium monnieri; A6: Coriandrum sativum; A7: Pastinaca sativa; A8: Heracleum candicans; A9: Foeniculum vulgare; A10: Ligusticum jeholense; A11: Nothosmyrnium japonicum; A12: Pimpinella diversifolia; A13: Cyclospermum leptophyllum; B1: Pternopetalum trichomanifolium; B2: Ostericum grosseserratum; B3: Anthriscus sylvestris; B4: Torilis scabra; B5: Daucus carota; C1: Oenanthe hookeri; C2: O. linearis; C3: Cryptotaenia japonica; D: Haplosphaera phaea; E: Hymenidium davidii; F: Bupleurum chinense; G: Chamaesium paradoxum; H: Sanicula lamelligera. A drawing of a transverse section through each fruit is displayed on the corresponding branch of the phylogenetic tree. An extra fruit transverse section ‘I’ shown at the intersection of two Angelica species (A1 and A2) belongs to A. dahurica.

Divergence time estimates compared with previous studies

The estimated times here from BEAST analyses using the concatenated dataset of 11 SCGs (Fig. 5) were similar to those based on ITS sequences (Banasiak et al., 2013) with deviations of c. 1–6 Myr (Table 3), except for the divergence time of tribe Chamaesieae, which is almost identical to those of the mean times assessed under a PL (penalized-likelihood) method based on rps16 sequences (Calviño et al., 2016), with a difference of only c. 1.62 Myr. The most recent common ancestor (MRCA) of Saniculoideae diverged from that of Apioideae at 56.64 Mya (95% HPD, 45.18–73.53 Mya) during the Late Cretaceous–Eocene, just c. 2.86 Myr later than estimated by Calviño et al. (2016) (Fig. 5; Table 3). This suggests the fossils we used to estimate the time of divergence in the Apioideae may be appropriate, contrary to the views of Calviño et al. (2016). Within the subfamily Apioideae, the MRCAs of Chamaesieae, Bupleureae and Pleurospermeae were estimated to have diverged from others during the Eocene at c. 49.78, 44.88 and 41.15 Mya, respectively. The MRCA of the Physospermopsis Clade later diverged from the distal branch D at 33.07 Mya (27.44–40.01 Mya), the divergence between Group C and the MRCA of Group A+B occurred 26.39 Mya (20.78–32.13 Mya), and Group A and Group B split from each other at 24.51 Mya (19.32–29.98 Mya). Within Group A, the MRCA of Careae and Pyramidoptereae separated from the MRCA of the remaining lineages at 21.38 Mya (16.8–26.5 Mya). All the major lineages of Group A originated and diversified between c. 21.24 and 7.99 Mya. The phylogeny position of the Sinodielsia Clade differs from the species tree; this lineage diverged from the MRCA of Tordylieae and Apieae at 13.21 Mya (9.65–17.07 Mya), and Tordylieae diverged from Apieae at 12.29 Mya (8.84–16.11 Mya). The subtribe Scandicinae, which is located as the most basal branch of Group B, diverged from the MRCA of the remaining lineages at c. 19.44 Mya (14.67–24.52 Mya); the separation of subtribes Daucinae and Torilidinae was during the middle Miocene, c. 12.01 Mya (8.21–16.66 Mya), and the sister groups split from the Acronema Clade at c. 14.52 Mya (10.61–19.41 Mya).

Fig. 5.

Chronogram presenting estimated divergence times by BEAST using 11 single copy genes and calibrated with Tertiary fossil pollen. Calibration points are marked with black circles. Blue bars and numbers above represent the 95 % highest posterior density (95% HPD) for each node. The scale axis is in millions of years ago (Mya). Below the scale axis shows the geological sequence of events related to the uplift of the QTP, including a graphical representation of the extent of the uplift through time. The top purple axis is the time period associated with QTP uplift and climate change (with reference to Faver et al., 2015).

Table 3.

Comparison of estimated divergence times (Mya) of major lineages; time was estimated by BEAST methods based on 11 concatenated single copy genes

Node	ITS	rps16	Single copy genes
Group A (Pyramidoptereae – Selineae)	26.1 (22.14, 30.4)	About 30.5	21.38 (16.8, 26.5)
Group B (Scandicinae – Acronema Clade)	25.06 (22, 28.7)	About 26.5	19.44 (14.67, 24.52)
Group A – Group B	About 31.5	About 33.5	24.51 (19.32, 29.98)
Group C (Oenantheae)	22.92 (19.16, 26.97)	About 30.5	12.38 (7.05, 19.18)
(Group A + Group B) – Group C	32.12 (28.41, 36.20)	About 36.5	26.39 (20.78, 32.13)
Physospermopsis Clade	21.7 (14.92, 30.06)	?	33.07 (27.44, 40.01)
Pleurospermeae	38.9 (35.97, 42.76)	About 47	41.15 (36.65, 47.8)
Bupleureae	44.51 (39.11, 51.55)	About 49.5	44.88 (39.49, 52.41)
Chamaesieae	20.25 (14.05, 27.61)	51.4 (37.2, 68)	49.78 (42.72, 58.7)
Saniculoideae – Apioideae	?	About 59.5	56.64 (45.18, 73.53)

Times estimated based on ITS refer to Banasiak et al. (2013) with three fossil calibrations; times estimated based on rps16 under a penalized-likelihood (PL) method refer to Calviño et al. (2016) with one fossil calibration; times estimated based on single copy genes are from our research with two fossil calibrations. Ranges correspond to 95% highest posterior density (95% HPD) or approximate bootstrap confidence quadratic (ABCq) confidence intervals. ‘?’ None available.

DISCUSSION

Rapid evolutionary divergence and inference of phylogenetic relationships among major lineages of Apioideae

The main aim of this study was to resolve and explain the long-standing controversy of relationships between the major lineages in Apioideae. As we have shown, the species trees constructed using MP-EST, STAR and ASTRAL-II all show a maximally supported topology [Chamaesieae, (Bupleureae, (Pleurospermeae, (Physospermopsis Clade, (Group C, (Group A, Group B)))))], with the positions of some major lineages differing from previous studies (Zhou et al., 2009; Banasiak et al., 2013; Calviño et al., 2016). The relationships among the basal lineages are similar to those inferred from cpDNA data only (Zhou et al., 2009; Calviño et al., 2016), and our placements of the Sinodielsia Clade and Apieae conflicted with those inferred from all of the published phylogenies (Zhou et al., 2009; Banasiak et al., 2013; Calviño et al., 2016). The very low support value (62/0.47) for branch III in the species tree of Group A (resampled datasets, Supplementary Data Fig. S2), and topologies of MP-EST species trees of Bt and Pt (with weak or moderate support, Figs S8, S15) conflict with those generated by other methods (STAR and ASTRAL-II), and the low quartet support values of branches III and IV regardless of the dataset used (Figs S1, S3, S9, S11, S13 and S16) suggest that the position of the Sinodielsia Clade and Apieae remains to be confirmed with high confidence. Our species tree strongly supports Cyclospermum letophyllum (Pyramidoptereae) as a sister group to Careae, the placement of which was variable in previous analyses based on ITS and rps16 intron (Banasiak et al., 2013; Calviño et al., 2016). The relationship among lineages in Group B is similar to that of Zhou et al. (2009) based on cpDNA, although sampling for both studies is incomplete and additional studies are needed.

Our analyses of the weighted datasets show that the species tree of the Pt dataset, with very high ILS level, shares the same topology with the final species tree, indicating that ASTRAL-II can reconstruct an accurate species tree in the context of a high ILS level. However, when the ILS level is low, the weighted datasets (Ot and Rt) may produce false species trees, indicating that in this situation the input gene trees have a large impact on the estimation of the species tree. This explains why we tend to get conflicting species trees based on small datasets, because such analyses may only yield particular trees in particular cases. Because this subfamily has a large number of species, when the species number is increased dramatically, existing species tree estimation methods, even ASTRAL-II, could not be run in a reasonable time. Thus, although weighted datasets appear to be unscientific for species trees estimation, we can use the weighted cluster method to reversely analyse and screening out a set of genes with similar tree topologies to the final species tree; these genes can be applied to the reconstruction of the phylogenetic framework of the subfamily Apioideae. This study provides a feasible method for the estimation of species trees in the subfamily Apioideae, which will be further discussed in future publications. Our results show that the topologies of gene trees for the Bt dataset should be the most similar to that of the final species tree, and thus we speculate that the corresponding genes of this dataset may play key roles in the whole evolutionary histories of Apioideae. It would be of value to explore the function of these genes, which may help to explain how the subfamily Apioideae has evolved.

In our ASTRAL-II analyses, branches I–V received one high-frequency topology and two lower frequency topologies (Fig. 2), as expected when the conflict between gene trees is caused by ILS, which is always a close companion of rapid evolutionary divergence (e.g. Maddison, 1997). The incongruence among these branches is unlikely to be explained using hybridization and introgression during their early evolutionary history. If the conflict between gene trees is caused by hybridization and introgression, one might expect two major (and equivalent) frequency gene tree topologies (e.g. if the branch was a result of hybrid speciation) or some other set of frequencies (e.g. if a sub-set of the genome introgressed at this point). Thus, the pattern of gene tree topology frequencies we found above is more consistent with the scenario of ILS than with hybridization and introgression (Mao et al., 2019). The relatively short branch lengths of these branches additionally indicate high levels of ILS (Mirarab and Warnow, 2015), consistent with rapid evolutionary divergence.

Fruit dispersal modes and character evolution of Apioideae based on the species tree

The diaspores of Apiaceae fruits are usually described based on their shape and appendages as anemochorous (dispersed by wind), epizoochorous (carried away on animal fur or feathers), hydrochorous (floating on water), or without any distinct adaptations to dispersal as barochorous (gravity-dispersed) (Wojewódzka et al., 2019). The various specializations of ribs and pericarp were conducive to seed dispersal (Fuentes and Vivian-Smith, 2009). In our study, fruits of Group C are clearly hydrochorous, with the corky mesocarp and waxy pericarp improving floating capabilities, indicating these fruits may favour long-distance dispersal by water. In some additional habitats (far from flowing water) they were barochorous in order to maintain the local population.

Fruit appendages facilitating wind dispersal may be of different origins, such as developing from either primary (for most taxa of Apioideae) or secondary ribs (within taxa of Daucinae, such as Thapsia and Laserpitium) (Weitzel et al., 2014; Banasiak et al., 2016; Wojewódzka et al., 2019). The only study on wind dispersal in winged-fruited Umbellifers concluded that its effectiveness as a mode of dispersal is quite low (Jongejans and Telenius, 2001). Other studies have suggested that bristles and hooks increase attachment and retention potential on animal fur (Römermann et al., 2005; Tackenberg et al., 2006) and promote gene flow among populations (Williams, 1994; Williams and Guries, 1994); recent research by Wojewódzka et al. (2019) suggested that epizoochorous fruits were generally dispersed at higher distances than winged fruits. These studies all rejected a role of anemochory in dispersal of Apioideae fruits. However, we reviewed these studies and suggest that the design of these studies might have ignored the influence of the environment on the survival of plants (habitat, wind frequency, type and population of local vectors, etc.), or were based on only a particular group (Wojewódzka et al., 2019), which was not sufficient to represent the fruit evolution characteristics of the entire subfamily. Our results have shown that the species with widely winged fruits generally live in high-altitude areas with an open environment, which would favour long-distance dispersal by air through flight (Theobald, 1971; Downie et al., 2002; Spalik et al., 2004; Calviño et al., 2008, 2016; Fernández et al., 2017a, b) and expand the dispersal distance by occupying more living space. We also suggest that characters such as fruit size, degradation of the dorsal and lateral primary ribs, and compression of the endosperm are an adaptation to this dispersal model. Winged fruits have been present throughout the evolution of Apioideae, even in the basal lineage Chamaesieae (C. mallaeanum) (Guo et al., 2018) and the most spiny fruit lineage Daucinae. As referred to in previous studies, D. decipiens and D. edulis are endemic to Madeira, where has never been connected to the mainland and lacks native large mammals that could have acted as dispersing agents for epizoochorous species, with the appendages of secondary ribs winged (Banasiak et al., 2016; Wojewódzka et al., 2019). This indicated that anemochory could be the best dispersal mode for species of Apioideae if animal vectors are absent, and the formation of wings for fruit ribs becomes an advantageous adaptation.

For most taxa of Daucinae and Torilidinae, epizoochory might be the best dispersal mode. The dispersal unit of these groups is not only individual mericarps but also the entire plant (for Daucus), which combined with slightly or strongly compressed endosperms of these fruits suggests an adaptation for anemochory. Such dispersal strategies may be used to explain the diversity and wide distribution of Daucus and the great differentiation seen within D. carota. Species with less developed ribs in Apioideae always have smaller fruits with a dense endosperm (Bupleureae, Chamaesieae and Pimpinelleae), which may be conducive to dispersal via gravity, and the smaller size and mass of these fruits also appear to be an adaptation to anemochory.

We mapped the fruit morphology and their dispersal modes into our species tree and speculated that the fruits of Apioideae might have undergone adaptive evolution and evolved in two directions. One is towards forming fruits such as those of Group C, which favour dispersal by water. The others tend to develop winged fruits, like fruits of Group A and Group B, adapted to dispersal by wind. The spiny fruits in Group B with spines evolved from wings developing on secondary ribs (Wojewódzka et al., 2019), suggesting epizoochory as a derived mode of anemochory. Interestingly, the fruit characters of Group A and the Acronema Clade are distinctly homoplastic, which may reflect convergent adaptive evolution among these lineages. However, when some exceptional cases, such as taxa of Pimpinelleae, are considered, the entire plant habit must also be taken into account. Normalized parameters are expected to be proposed for the dispersal of Apioideae in the future, which could integrate most of the characters that impact fruit dispersal. These characters, as we suggest, should include fruit size, development of ribs, habitat, type and population of local vectors, and so on. All of these together determine the efficiency and modes of fruit dispersal.

An updated evolutionary divergence time scale of Apioideae related to uplift of the QTP

Our divergence time estimate is credible because it depends on a reliable phylogeny and displays short 95% HPDs. The 95% HPDs on ages for our selected nodes are less than 13 Myr except the time of origin of Chamaesieae and Apioideae. Nevertheless, our 95% HPD for the time of origin of Chamaesieae is 15.98 Myr (42.72–58.7 Mya), about half that inferred from rps16 (30.8 Myr) (Table 3) (Calviño et al., 2016) and shares part of the time scale in that study (37.2–68 Mya). Additionally, long 95% HPDs (some more than 30 Myr) are given for the six selected nodes of Apioideae lineages estimated by Calviño et al. (2016), which were much longer than our results. Even though 95% HPDs estimated by ITS were almost all shorter than those from our study, the uncertain phylogeny restricted their accuracy. We speculated that the evolution of Apioideae might be related to uplift of the Qinghai–Tibetan Plateau (QTP) based on this credible estimation of divergence time. Taxa of Apioideae are distributed widely in the North Temperate zone, and in East Asia, China is recognized as one of the most important distribution centres. The main centre of diversity of Apioideae is south-west China, including Sichuan, Yunnan and adjacent parts of Xizang A. R (Pimenov, 2017), across the QTP and the Hengduan Mountains. We adopted the geological scenario related to uplift of the QTP from Favre et al. (2015) (Fig. 5). It is undoubtedly very attractive to explore the complex evolution of Apioideae, a group that evolved over almost the entire period of the orogeny, probably closely related to these geological events. Our estimated divergence time of Apioideae (Fig. 5) suggested that Apioideae originated at c. 56.64 Mya (95% HPD, 45.18–73.53 Mya) slightly earlier than the collision of the Indian–Eurasian plates (van Hinsbergen et al., 2012; Favre et al., 2015). The divergence time of two readily identifiable earliest diverging lineages (Chamaesieae and Bupleureae) of Apioideae in Asia probably indicates that Indian–Eurasian collision did not cause a sharp differentiation of Apioideae taxa. An alpine lineage (Pleurospermeae) originated at c. 41.15 Mya (95% HPD, 36.65–47.8 Mya), consistent with the high elevation for the QTP during the early period of uplift (Favre et al., 2015; Renner, 2016). The formation time of the morphologically rich taxa of distal branch D and the Physospermopsis Clade was c. 33.07 Mya (95% HPD, 27.44–40.01 Mya), suggesting that the divergence of their MRCA was occurred during time of uplift of the QTP. The divergence time of the three successive sister groups (A, B, C) within branch D with deviations of only c. 1.88 Myr indicates that the ancestors of these species might have undergone rapid evolutionary divergence, consistent with our species tree with short branch length and low quartet support (q1 = 0.49, q2 = 0.22, q3 = 0.3). The major clades of these three groups (A, B, C) generally originated at 11.3–21.38 Mya, possibly related not only to uplift but also to climate changes (Favre et al., 2015). The evolution of some Apioideae genera might have been driven by the final period of uplift (Favre et al., 2015), as inferred from the divergence times (1.3–7.44 Mya) within genera (Angelica, Pternopetalum and Oenanthe), mainly during the orogeny of the Hengduan Mountains and Qaidam basin. The origin of the two most controversial lineages in our studies, Apieae (c. 12.29 Mya) and the Sinodielsia Clade (c. 13.21 Mya), is relatively short (c. 0.92 Myr), consistent with rapid evolutionary change inferred from our species tree. The Acronema Clade and Physospermopsis Clade, which are almost exclusively found in east Asian (Zhou et al., 2008; Zhou et al., 2009) and vary widely in morphological characters, were suggested to have originated and diverged in close association with the uplift of the QTP (Fig. 5) and might have undergone rapid evolutionary divergence. Overall, our studies indicate that the QTP was likely to be an origin and differentiation centre of Apioideae, and the taxa of these two east Asian endemic clades (Acronema Clade and Physospermopsis Clade) might provide details regarding the significance of the QTP orogeny on the evolution of Apioideae. The significant convergence of previously considered important morphological characters indicated that taxa of Apioideae living in these regions might be likely to undergo rapid radiations. Climate models modified by orogeny are crucial to the biodiversity of these regions, and the time frame of the orogeny provides clues to the evolution of Apioideae. Fruit and dating analyses might reflect a convergent evolution between the Acronema Clade and Group A. Unfortunately, we did not study the biogeography of Apioideae in depth due to the small number of groups selected. This is expected to be completed in future. Combining divergence time and the relevant geological events, we infer that the orogeny of the QTP might have triggered the diversification of Apioideae, most taxa of which were still on the path of evolution, helping us to understand the complexity of the internal classification of Apioideae.

CONCLUSION

Phylogenetic relationships among major lineages of Apioideae have been a contentious issue due to the different molecular markers used. Our study based on 3351 SCGs yielded a reliable species tree with five branches with low quartet support values, which suggested that rapid evolutionary divergence and ILS may have been the main cause of conflicts observed among gene trees. Our species tree differs from the phylogenies inferred from ITS and cpDNA (Zhou et al., 2009; Banasiak et al., 2013; Calviño et al., 2016), with rearrangements among some major lineages. Both our phylogeny and fruit analyses strongly support that Scandiceae should be divided into at least two major clades. One includes Daucinae and Torilidinae, which should be restored as tribes according to both molecular and morphological evidence. The presence of fruit wings seems to be an optimization of Apioideae fruits over the present phylogenies, which gave the group an adaptive advantage that allowed it to diversify rapidly. The small number of groups may cause some deviation in the divergence time estimates, although it is credible to some degree in the context of a reliable phylogeny and short 95% HPDs. Molecular dating analysis based on 11 concatenated SCGs suggested that Apioideae originated around 56.64 Mya (95 % HPD, 45.18–73.53 Mya), and the QTP region acted as a probable diversity centre of Apioideae. Uplift of the QTP and climatic changes probably drove rapid speciation and diversification of Apioideae in the QTP region. Our study shows that combining SCGs and coalescent-based species tree estimation methods is a powerful approach that provides more refined phylogenetic estimates for Apioideae that were controversial based on small datasets.

SUPPLEMENTARY DATA

Supplementary data are available online at https://academic.oup.com/aob and consist of the following.

Fig. S1: Quartet support for each branch in the species tree of Apioideae.

Fig. S2: Species tree of Group A estimated based on 3793 gene trees using ASTRAL-II.

Fig. S3: Quartet support for each branches in the species tree of Group A.

Fig. S4: Species tree of Group B estimated based on 4257 gene trees using ASTRAL-II.

Fig. S5: Quartet support for each branch in the species tree of Group B.

Fig. S6: A cluster of all the 3351 gene trees upon neighbour-joining analysis.

Fig. S7: Species tree of the Bt branch generated using ASTRAL-II based on 1184 SCG trees.

Fig. S8: Species tree of the Bt branch generated using MP-EST based on 1184 SCG trees.

Fig. S9: Quartet support for each branch in the species tree of the Bt branch.

Fig. S10: Species tree of the Ot branch generated using ASTRAL-II based on 1260 SCG trees.

Fig. S11: Quartet support for each branch in the species tree of the Ot branch.

Fig. S12: Species tree of the Rt branch generated using ASTRAL-II based on 405 SCG trees.

Fig. S13: Quartet support for each branch in the species tree of the Rt branch.

Fig. S14: Species tree of the Pt branch generated using ASTRAL-II based on 502 SCG trees.

Fig. S15: Species tree of the Pt branch generated using MP-EST based on 502 SCG trees.

Fig. S16: Quartet support for each branch in the species tree of the Pt branch.

Fig. S17: Reconstruction of the two selected fruit characters on the final species tree.

Table S1: Collection records of the 27 transcriptome datasets sample.

Table S2: Final quartet score and normalized quartet score for the datasets we explored.

FUNDING

This work was supported by the National Natural Science Foundation of China (Grant Nos. 31872647), the fourth national survey of traditional Chinese medicine resources (Grant No. 2019PC002), and the National Infrastructure of Natural Resources for Science and Technology (Grant No. 2005DKA21403-JK).