Phylogeny of the Genus Chrysanthemum L.: Evidence from Single-Copy Nuclear Gene and Chloroplast DNA Sequences

Chrysanthemum L. (Asteraceae-Anthemideae) is a genus with rapid speciation. It comprises about 40 species, most of which are distributed in East Asia. Many of these are narrowly distributed and habitat-specific. Considerable variations in morphology and ploidy are found in this genus. Some species have been the subjects of many studies, but the relationships between Chrysanthemum and its allies and the phylogeny of this genus remain poorly understood. In the present study, 32 species/varieties from Chrysanthemum and 11 from the allied genera were analyzed using DNA sequences of the single-copy nuclear CDS gene and seven cpDNA loci (psbA-trnH, trnC-ycf6, ycf6-psbM, trnY-rpoB, rpS4-trnT, trnL-F, and rpL16). The cpDNA and nuclear CDS gene trees both suggest that 1) Chrysanthemum is not a monophyletic taxon, and the affinity between Chrysanthemum and Ajania is so close that these two genera should be incorporated taxonomically; 2) Phaeostigma is more closely related to the Chrysanthemum+Ajania than other generic allies. According to pollen morphology and to the present cpDNA and CDS data, Ajania purpurea is a member of Phaeostigma. Species differentiation in Chrysanthemum appears to be correlated with geographic and environmental conditions. The Chinese Chrysanthemum species can be divided into two groups, the C. zawadskii group and the C. indicum group. The former is distributed in northern China and the latter in southern China. Many polyploid species, such as C. argyrophyllum, may have originated from allopolyploidization involving divergent progenitors. Considering all the evidence from present and previous studies, we conclude that geographic and ecological factors as well as hybridization and polyploidy play important roles in the divergence and speciation of the genus Chrysanthemum.


Introduction
Chrysanthemum L. (Asteraceae-Anthemideae) is a genus associated with polyploidy and hybridization [1][2][3][4][5]. It is well known for its commercial chrysanthemum cultivars, which are globally important cut flowers and pot plants [6,7]. This genus comprises about 40 species mainly distributed in East Asia, and the species diversity is centered in China [8][9][10]. The considerable variations in morphology and ploidy level are exhibited in this genus (from 2n = 2x = 18, to 2n = 36, 54, 72, up to 90) [1,11]. Infraspecies and even infrapopulation variations in ploidy have been found in some cases [1,2,12]. In addition, many species of Chrysanthemum are narrowly distributed and habitat-specific [1,4,8,9]. All this information indicates that speciation is associated with hybridization, polyploidy, and adaptation. So far, the evolutionary history of Chrysanthemum has not been well understood despite a lot of efforts [13,14].
There is a great deal of controversy regarding its scientific name, entity and circumscription [15][16][17]. This situation has been improved through the resurrection of Chrysanthemum L. for the Asiatic species formerly under Dendranthema (DC.) Des Moul., and of Glebionis Cass. for the two Mediterranean species, C. segetum and C. coronarium, which had formerly been described under Chrysanthemum [8,9,18]. Because of the economic importance of garden chrysanthemums, an impressive amount of literature has accumulated on their breeding, cultivation, and origins [6,7]. However, the relationships between the wild Chrysanthemum species remain poorly understood.
As shown in the morphology-based cladogram, Chrysanthemum has close relationships with Brachanthemum, Arctanthemum, Tridactylina, Ajania, and Phaeostigma [9]. It is distinguished from them by its obovoid, thin-walled, myxogenic cypselas without pappus, and involucral bracts with dark brown margins [9]. However, the relationships based on the morphological characters show little congruence with findings from molecular data [18][19][20]. In the molecular phylogeny of the subtribe Artemisiinae, there are two main groups, the Artemisia-group and the Chrysanthemum-group ( = Dendranthema group) [19]. The latter, a strongly supported clade in the ITS tree, consists of Chrysanthemum, Ajania, Arctanthemum, and Elachanthemum [18,19]. Moreover, the taxonomic positions of Ajania and Phaeostigma relative to Chrysanthemum have long been controversial due to the pollen exine ornamentations and other morphological traits, such as the type of ray floret [9,21,22].
Ajania and Chrysanthemum have the same, common type of pollen grains, Anthemis-type pollen with echinate, while the representative of Phaeostigma has Artemisia-type pollen with microechinate, measuring ,1 mm in height [20][21][22]. Ajania and Phaeostigma shared the common disciform capitula, while Chrysanthemum has the radiate capitula [9]. Bremer and Humphries proposed that Phaeostigma might be a grade between Ajania and Artemisia, and suggested that circumscribing Phaeostigma requires more extensive study covering greater numbers of Ajania species [9]. Ajania purpurea, for example, is a small perennial only found in alpine habitats at elevations of 4800-5300 m, Xizang (Tibet) [8]. Based on our field observations, brownish style-branches and purple corolla with erect lobes distinguish it from other Ajania species. The phylogenetic position of this species needs to be clarified.
Interspecific relationships within Chrysanthemum are complicated because hybridization and polyploidy are common in this genus [1][2][3][4][5]. Although molecular phylogenetic data have been accumulated in the Chrysanthemum species, most of these data covered a small geographical range and focused primarily on garden cultivars and a few closely related wild species [23][24][25]. Two recent studies with larger numbers of samples and more molecular markers improved the phylogeny reconstruction of Chrysanthemum [13,14]. However, the interspecific relationships still remain undetermined due to the limited informative sites of nuclear ribosomal and chloroplast DNA data. We here try to elucidate the evolutionary history of Chrysanthemum using DNA sequence data containing relatively large amounts of phylogenetic information from both nuclear and chloroplast genomes.
In terms of resolving the phylogeny of the taxa, particularly those with rapid speciation and complex relationships, single-or low-copy nuclear genes are superior to organelle and nuclear ribosomal DNA [26,27]. First, they generally have high levels of variation to infer relevant gene trees with confidence. Second, they contain signals of hybridization and/or introgression if homeologous alleles are present in an individual or population. The analyses of single-or low-copy nuclear genes have helped with untangling the complex history of such taxa as Cerastium, Paeonia, Oxalis, and Achillea [28][29][30][31].
One candidate nuclear gene for this study is the chrysanthemyl diphosphate synthase gene (CDS). CDS catalyzes the condensation of two molecules of dimethylallyl diphosphate to chrysanthemyl diphosphate, an irregular monoterpene [32,33]. Irregular monoterpenes are mainly found in plants of the tribe Anthemideae of Asteraceae [32][33][34]. CDS is a member of the farnesyl diphosphate synthase (FDS) gene family, but it differs from other members of the family by an N-terminal extension of 50 amino acids, which has been identified as a plastidial targeting peptide [32,33]. Wang et al. isolated a fragment of CDS from Chrysanthemum 6 morifolium [35]. Their data were used to explore relationships among garden chrysanthemums and their affinities. However, that study provided little phylogenetic information due to the scarcity of samples from the wild species. Meanwhile, we have studied molecular evolution of the FDS gene family (Liu et al. unpublished). This allows us to identify CDS orthologs among the Chrysanthemum species and find variation-rich regions for the present study.
Based on a broad species sampling, we here conduct a phylogenetic study using DNA sequences of the nuclear singlecopy CDS gene and seven cpDNA loci (psbA-trnH, trnC-ycf6, ycf6-psbM, trnY-rpoB, rpS4-trnT, trnL-F, and rpL16 intron). Our major objectives are 1) to investigate the relationships between Chrysanthemum and its generic allies in the subtribe Artemisiinae, 2) to examine the interspecific relationships within Chrysanthemum, and 3) to verify polyploidy and hybridization speciation events common in Chrysanthemum.

Materials and Methods
Forty-three species/varieties of Chrysanthemum and of seven other genera of the subtribe Artemisiinae (Anthemideae, Asteraceae) were sampled. Among them, 32 were from Chrysanthemum, four Ajania, two Phaeostigma, and one each from Elachanthemum, Artemisia, Brachanthemum, Kaschgaria, and Stilpnolepis, respectively (Table 1). For the diploid species and the narrowly distributed polyploid species, one individual from every studied population was sampled. For species with varying ploidy levels and a broad distribution, such as C. indicum, C. naktongens, and C. zawadskii, 2-5 populations were sampled with one individual from each population. For species with multiple varieties, one individual per variety studied was sampled. To incorporate information from garden chrysanthemums, two cultivated varieties of C. morifolium (accessions ''ZY'' and ''HSZ'') representing two flower color series, yellow and white-purple, were included in this study (Table 1). Fresh leaves were collected in the field, desiccated, and stored in silica gel. The same plant individuals were sequenced at all the DNA loci. All the samples used in this study were collected by the senior author and his colleagues or cooperators, and no special permissions were required for collection. The locations are not privately owned or protected in any way and the field studies do not involve endangered or protected species. Voucher specimens were deposited in the Peking University herbarium (PEY).

Ploidy level examination
The DNA ploidy level of each individual sample was checked by means of flow cytometry using leaves dried in silica gel [36,37]. To produce a standard control for the DNA ploidy level determination, traditional chromosome counting was conducted from root tips of a diploid sample of Chrysanthemum chanetii. Root tips were pretreated in a saturated water solution of alpha-bromonaphthalene for 2 h and then fixed in a solution of ethanol: acetic acid (3:1) for one day. Chromosomes were stained with acetocarmine and counted under a microscope.

DNA extraction, amplification, and sequencing
Total genomic DNA was isolated from silica-gel dried leaves using the 26 CTAB method [38] or using the Plant Genomic DNA Kit (TianGen Biotech, Beijing, China).
The nuclear chrysanthemyl diphosphate synthase (CDS) gene was amplified, cloned and sequenced. To verify the single-copy property of the CDS gene, its entire length (containing 13 exons and 12 introns) was sequenced from Chrysanthemum lavandulifolium and several species from other genera within Asteraceae. An Nterminal amino acid extension of 50 amino acids allowed the distinction of CDS from its paralogs (Liu et al. unpublished). For the present phylogenetic analysis, the region from exon 5 through 10 was sequenced using specific primers CDSII, CDSVb, CDSVa [35] and CDSIIa (designed by this study). The amplification was conducted first with the forward primer CDSII and the reverse CDSVb. This was followed by a nested PCR using the forward CDSIIa and the reverse CDSVa ( Figure 1). Detailed information regarding the primers is given in Table 2.
The polymerase chain reaction was performed in a total volume of 20 mL containing 2 mL 106 buffer, 0.5 mM of each primer, 200 mM of each dNTP, 1 U of exTaq (TaKaRa, Shiga, Japan) or Taq polymerase (TianGen Biotech, Beijing, China), and 1 mL template genomic DNA. The amplification followed a program of 5 min at 95uC, 35 cycles of 1 min at 95uC, 1 min at 56uC, and 1 min at 72uC, and a final extension of 5 min at 72uC. The PCR products were excised from the 0.8% agarose gel and purified with a Gel Extraction Mini Kit (Watson Biotech, Shanghai, China). Direct sequencing was conducted for the chloroplast DNA fragments using the same primers as in the amplification. To sequence the nuclear CDS gene, the purified PCR products were ligated into a pGEM-T Easy Vector using a Promega Kit (Promega Corporation, Madison, WI, U.S.). Plasmids containing the right insertion were chosen for the sequencing reaction, and basically five clones were sequenced per individual. Cycle sequencing was conducted using the vector-specific universal T7 and SP6 primers, and the products were run on a 3730 automatic DNA sequencer. All sequences analyzed were deposited in GenBank (Accession Numbers JF939945-JF940511; JQ612544).

Data analysis
Sequences were aligned with ClustalX version 1.81 [41] and then visually checked and modified with BioEdit version 7.0.1. Possible PCR-recombinant sequences [42][43][44] of the nuclear CDS were identified using method described by Russel et al. [45] and eliminated.
Because polyploids are common in Chrysanthemum and some of them are probably of hybrid origin, the resolution and topology of the phylogenetic trees may be interfered by the hybrid-polyploids [46]. To minimize this influence, we analyzed the present data in two steps: First, we focused on the circumscription and monophyly of Chrysanthemum and its relationships with the allied genera by excluding all the polyploid Chrysanthemum samples (Figures 2A and  2B). The monotypic genus Stilpnolepis at the relative basal position of phylogeny of the subtribe Artemisiinae [18,19] was defined as the outgroup. Because the intron sequences of the CDS gene are hypervariable at the generic level and cannot be properly aligned across all the genera studied here, the CDS gene tree of the diploid Chrysanthemum and its allied genera ( Figure 2B) was constructed using only their exon sequences. In the second step, we focused on the phylogeny of Chrysanthemum and all the Chrysanthemum species were included. Stilpnolepis was still used to root the cpDNA tree in this analysis. To obtain a better solution of relationships within Chrysanthemum, CDS intron sequences were also used. Because the Table 2. Primers used for this study.
PLOS ONE | www.plosone.org CDS intron sequences of the Chrysanthemum species can only be aligned with those of Ajania and Phaeostigma, but not with any putative outgoups, samples of these two genera together with Chrysanthemum were analyzed and an unrooted tree was built on the the CDS exon+intron sequences. Incongruence length difference (ILD) test [47] implemented in PAUP 4.0b10 [48] was used to assess the congruence of the data, between each pair of the seven chloroplast loci, between the exons and introns of CDS, and between the combined cpDNA and nuclear CDS data. When the test was conducted between chloroplast and nuclear DNA, only plant individuals lacking divergent nuclear haplotypes were used. Because CDS intron sequences only from the Chrysanthemum, Ajania, and Phaeostigma species could be aligned, the exons and introns from these samples were used to perform the ILD test. The test was performed with 1000 replicates, random taxon addition (10 replicates) and a maximum of 1000 trees saved per round. The results showed no significant incongruence between any pairs of the cpDNA loci (p ranges from 0.197 to 1.000) or between exons and introns of CDS (P = 0.884), but there was significant ionncgruence between the combined cpDNA and the nuclear CDS datasets (P = 0.001). The seven cpDNA loci were then combined for the analysis, and the exons and introns of CDS from Chrysanthemum, Ajania, and Phaeostigma were also combined for analysis. The chloroplast and nuclear data were analyzed separately.
Phylogenetic analyses were conducted using Bayesian inference (BI) and maximum parsimony (MP) methods. In all analyses, gaps were treated as missing data.
For the Bayesian analyses, the best-fit nucleotide substitution models were selected using Akaike information criterion in MODELTEST 3.06 [49]: TVM+I and K81uf+G for the cpDNA and CDS exon data sets (with the diploid Chrysanthemum species and  all the allied genera), and GTR+I+G and K81uf+G for the cpDNA (covering all the samples) and CDS intron + exon data sets (covering all the Chrysanthemum species and Ajania and Phaeostigma), respectively. Bayesian inference (BI) was conducted in MRBAYES 3.1.2 [50]. Two independent MCMC runs, each with four chains (three hot, one cold) were run simultaneously starting from a random tree for 10 million generations with one tree sampled every 1000 generations. According to stationarity of lnL assessed using Tracer v. 1.4 [51], the first 10% sampled trees prior to stationarity were discarded as burn in for all analyses, and the remaining trees were used to construct the 50% majority-rule consensus tree.
Maximum parsimony analyses were performed with PAUP 4.0b10 [48]. Heuristic search was performed with random taxon addition with 1000 replicates, TBR branch swapping, and saving multrees (no more than 10 trees with the score no less than 10 were saved per replicate). The MP bootstrap analysis was conducted with 1000 replicates, simple sequence addition, TBR branch swapping, and multree option on.

Examination of pollen
To examine the types of pollen variation among these genera, pollen grains of Phaeostigma salicifolium, P. variifolium, Ajania purpurea, A. przewalskii, Chrysanthemum indicum, and C. naktongense were collected from the voucher specimens used for the molecular analysis. Pollen grains were mounted on a metal stub, critical-point dried, and sputter-coated with gold. Observations and micrographs were performed under scanning electron microscopy (SEM, Hitachi S-4800) at 10 kV.

Variations in ploidy level
The results of flow cytometry are listed in Table 1. Considerable variation in ploidy level was found within Chrysanthemum. Ploidy ranged from 2x to 10x. Infra-species variation was also observed. For instance, C. indicum was found to have 2x (SNJ), 4x (NJ, SNJ, and GZ), and 6x (JP) populations, and C. zawadskii was found to have 4x (RS) and 6x (AR) populations. In most cases, the diploid species appear to be endemic or narrowly distributed, and polyploids are usually widely distributed.

Sequence characteristics
Seven chloroplast loci were each sequenced from 49 individuals of 43 taxa ( Table 1). The combined cpDNA data set has 3971 characters, of which 227 are variable and 70 parsimonyinformative. Of the seven plastid loci, the intergenic spacer psbA-trnH showed the highest percentage of informative characters as 4.90%, whereas trnY-rpoB the lowest as 0.84% (Table 3). Partial sequences of the nuclear CDS gene from its exon 5 through to exon 10 were sequenced. In total, 228 sequences were obtained from 49 individuals of 43 taxa (Table 3). These sequences fall into 150 haplotypes with a length variation from 879 bps (in Artemisia annua) to 1265 bps (in Chrysanthemum sichotensis). The aligned CDS exon data set covering all the studied taxa contains 429 nucleotide positions, of which 192 (44.76%) are variable and 184 (42.89%) parsimony informative ( Table 3). The CDS exon + intron data set covering Chrysanthemum, Ajania, and Phaeostigma contains 1374 nucleotide positions, of which 418 (30.42%) are variable and 378 (27.51%) are parsimony informative (Table 3). Circumscription of Chrysanthemum and its relationships with the generic allies To assess the circumscription of Chrysanthemum, only the diploid species of the genus and all its generic allies were included in the first step of analyses. The Bayesian inference and the maximum parsimony (MP) analyses generated almost the same topologies. The results of the MP analysis are only shown by bootstrap values mapped on the Bayesian trees (Figures 2A and 2B).
The close relationships of Chrysanthemum and Ajania are apparent in both the cpDNA and the CDS exon trees. Most Chrysanthemum and Ajania species formed a clade in the CDS exon tree (marked as w in Figure 2B), but C. arisanense, C. naktongense (population ZJ), C. sinuatum and A. purpurea were not included in this clade. In the cpDNA tree (Figure 2A), similar relationships were found except that Elachanthemum falls into the clade otherwise only of Chrysanthemum+Ajania (clade w in Figure 2A). A. purpurea, a tetraploid species, was grouped with two species of Phaeostigma in both trees (clades q in Figure 2A & B). In the CDS exon tree, Elachanthemum, Kaschagaria and Brachanthemum formed a clade sister to that consisting of Chrysanthemum, Ajania and Phaeostigma, while the intergeneric relationships were not resolved by the cpDNA data, i.e., Elachanthemum, Kaschagaria and Artemisia were intermingled with the Chrysanthemum+Ajania group ( Figure 2A). Therefore, Elachanthemum, Kaschagaria, and Artemisia were more close to Chrysanthemum than Phaeostigma and Brachanthemum do in the cpDNA tree, but the CDS gene tree suggests that Phaeostigma may be more closely related to Chrysanthemum and Ajania than all the other genera studied. The close relationship between Phaeostigma and the Chrysanthemum+Ajania is further supported by the fact that, among all the genera studied here, only Phaeostigma has CDS intron sequences that can be properly aligned with those of Chrysanthemum and Ajania.

Phylogeny of Chrysanthemum
Phylogenetic trees including polyploid Chrysanthemum species are shown in Figures 3 and 4 (an unrooted CDS intron+exon tree covering Chrysanthemum+Ajania+Phaeostigma).
In the cpDNA tree (Figure 3), most Chrysanthemum species fell into clade II. In this core group of Chrysanthemum, we recognize two major clades, III and IV. All the species in clade III are from China, and most of them have yellow ray florets and belong to the C. indicum group (C. rhombifolium and C. vestitum do not). In contrast, most of the species/varieties in clade IV have white-purple ray florets (C. boreale, C. dichrum, C. indicum (4x+6x), and C. okiense do not). In clade IV, subclade IVa includes all the species/varieties occurring in northern and northeastern China, Russia, the Korean Peninsula, and Japan. Figure 4 is based on the CDS exon + intron sequences. As shown in Figures 2A, 2B, and 3, the three diploid species C. sinuatum, C. naktongense (ZJ), and C. arisanense show rather distant relationships with other species within Chrysanthemum. Again, we find that Chrysanthemum and Ajania cannot be separated. Four major clades, I, II, III, and IV can be recognized for the majority of the Chrysanthemum species and some of the Ajania species (Figure 4). Clade I broadly corresponds to the C. indicum group (characterized by small capitula with yellow ray florets) except for three haplotypes of C. argyrophyllum. Clade II contains species with white-purple ray florets. These belong to the C. zawadskii group except for two haplotypes of C. indicum (NJ) and two haplotypes of C. argyrophyllum. Clade III includes Chrysanthemum species with white-purple ray florets and two Ajania species. Except for the octoploid C. ornatum, Clade III contains diploid and tetraploid species homozygous at the CDS locus. Clade IV exhibits considerable variation with respect to i) ploidy levels from 2x to 4x, 6x, 8x, and 10x, ii) yellow and white-purple ray florets, and iii) a broad geographic distribution.
Many of the polyploid samples harbor divergent nuclear CDS gene haplotypes, which indicate their hybrid origins. For example, the tetraploid C. indicum (NJ) has its CDS haplotypes scattered in clades I, II, and IVb; and the hexaploid C. indicum (JP) harbors divergent CDS haplotypes found in I and IVb; the hexaploid C. argyrophyllum has haplotypes in clade I and II; and the hexaploid C. vestitum has two haplotypes close to clade III and three in clade IVa.

Variations in pollen types
The six species examined here have been found to have two types of pollen exine ornamentations. They are the Anthemis-type (echinate) for Chrysanthemum indicum, C. naktongense, and Ajania przewalskii and the Artemisia-type (microechinate) for Phaeostigma salicifolium, P. variifolium, and A. purpurea ( Figure 5). For P. variifolium and C. naktongense, this is a first report of their pollen morphology. The close relationship between Ajania purpurea and Phaeostigma, reflected by pollen morphology, is in line with that indicated by DNA sequence data (Figures 2-4).

Discussion
Previous studies based on nrITS, nrETS, and cpDNA data have provided insights into the phylogeny of the tribe Anthemideae [13,14,[18][19][20]. They have also helped to resolve generic delimitation and circumscription of several groups within the subtribe Artemisiinae [13,14,[18][19][20]. However, as Masuda et al. have pointed out, species of Chrysanthemum and its allies may have undergone rapid adaptive diversification so that they appear to be unresolved by using DNA markers (nrITS, nrETS and cptrnL-F) that have relatively low levels of divergence [14].
Our data show that, for Chrysanthemum and its allies, the chrysanthemyl diphosphate synthase (CDS) gene harbors rich information at both the generic (with its exon regions) and species (mainly with the highly variable intron regions) levels. It contains substantially more polymorphic sites than the combined cpDNA loci (Table 3) and the nuclear ribosomal ITS regions [13]. Together with the data from seven chloroplast DNA loci, which are of relatively rich in potentially informative characters (PICs) across the angiosperms and in Asteraceae [39,40], the CDS gene sequences have helped us to develop a better understanding of the evolutionary history of Chrysanthemum and its allied genera.

Monophyly, circumscription, and close relatives of Chrysanthemum
Both the cpDNA and nuclear gene trees (Figure 2A & 2B) failed to support Chrysanthemum as a monophyletic genus. The affinity of Chrysanthemum with Ajania indicated by the present study is in line with the results of several other studies based on the nrDNA ITS, ETS, and cpDNA data [13,14,[18][19][20]. The genus Ajania was established by Poljakov based on its corymbose synflorescence of the disciform capitula [52]. He speculated that Ajania is a very old group derived from the same ancestors as Artemisia [52]. However, Tzvelev assumed a close affinity between Ajania and Chrysanthemum [53]. Masuda et al. suggested that Chrysanthemum (except for C. sinuatum) and Ajania were too close to be separated and that they should be treated within the genus Chrysanthemum sensu stricto [14]. According to our data, we agree with Masuda et al. that Ajania should be incorporated into Chrysanthemum despite their differences with respect to capitulum characters [14]. We also agree with Tzvelev that Ajania could have evolved from an ancestral taxon closely related to Chrysanthemum [53]. There are three types of capitula in the subtribe Artemisiinae: discoid, disciform, and radiate. According to recent studies, transcription factor CYCLOI-DEA-like (CYC-like) genes play an important role in the morphogenesis of flower symmetry [54][55][56]. The disciform capitula of the Ajania species might be due to convergent genetic or epigenetic changes related to functions of the CYC-like genes.
With respect to the relationships between Chrysanthemum+Ajania and other allied genera, incongruence is evident between the cpDNA and nuclear CDS data: In the cpDNA tree, Elachanthemum, Kaschgaria, and Artemisia appear to be more close to Chrysanthemu-m+Ajania than Phaeostigma and Brachanthemum (Figure 2A). According the nuclear CDS data, however, Phaeostigma is more closely related to Chrysanthemum+Ajania than any other genera ( Figure 2B). These conflicting relationships were also reported by other studies. The monophyly of the Chrysanthemum group ( = Dendranthema-group) including Chrysanthemum, Ajania and Elachanthemum was evident in several studies [13,[18][19][20]. However, the close affinity of Phaeostigma with Chrysanthemum+Ajania, especially with Ajania, was supported by the nuclear ribosomal ITS and ETS data and the capitulum traits [9,14].
The incongruence between the plastid and nuclear gene trees can be attributed to hybridization, incomplete lineage sorting, and different evolutionary rates and history of plastid and nuclear genomes or DNA regions, [57,58]. These factors make it difficult to interpret gene trees with any certainty. The major problems are the differentiation of speciation genes, discrepancy between phenotypic traits and the randomly selected DNA regions used for the phylogenetic survey, different re-combinations of parental genomes, and different patterns of lineage sorting during subsequent species radiations.
Retention of ancestral polymorphisms, or incomplete lineage sorting, is common in recently diverged species [57]. Relevant problems have been evident in many recently radiated groups of the Asteraceae, such as Tragopogon and Achillea [59][60][61]. In Achillea, by comparing the inferred species tree and the nuclear gene trees, Guo et al. demonstrated that lack of sorting of ancestral alleles exists not only among closely related species but extends to the taxa with relatively distant relationships [61]. In the present study, the incongruence between the plastid and nuclear CDS trees already on the 2x-level in Chrysanthemum and its generic allies mainly stem from lack of lineage sorting and/or hybridization. Hybridization within Chrysanthemum will be discuss in a later chapter. Here, we interpret the inter-generic relationships given the available molecular and morphological data.
Considering that i) plastid DNA variation is often too low to support the inference of species relationships, ii) Phaeostigma is the only genus among the allied genera that can be compared to Chrysanthemum + Ajania in terms of CDS intron sequences, and iii) Phaeostigma and Ajania both have disciform-heterogamous capitula, we conclude that Phaeostigma is more closely relative of the Chrysanthemum + Ajania group than Elachanthemum.
Phaeostigma is a small genus consisting of only three species. Muldashev separated it from Ajania on the basis of its brownish style-branches, erect corolla lobes and pollen characters [62]. According to Muldashev, Phaeostigma is closely related to the ancestors of Ajania and distantly related to Artemisia even though it has the same pollen morphology as the latter [62]. There are two types of pollen exine ornamentations in the subtribe Artemisiinae [20][21][22]. Phaeostigma has Artemisia-type pollen, and Ajania has Anthemis-type pollen [9,21]. Bremer and Humphries proposed that Phaeostigma might be a grade between Ajania and Artemisia [9]. They also suggested that circumscribing Phaeostigma requires more extensive study on greater numbers of Ajania species, especially those with habits similar to those of Phaeostigma. Our field observations showed that Phaeostigma plants have larger leaves and grow in more humid habitats than most Ajania species.
The distinction between Phaeostigma and Ajania seems to be difficult due to the placement of Ajania purpurea in the clade of Phaeostigma, which is placed with confidence by both gene trees (Figures 2A & 2B). Pollen traits show that A. purpurea shares with Phaeostigma the Artemisa-type, unlike the Anthemis-type of A. przewalskii ( Figure 5) and other Ajania species [20][21][22]. A. purpurea is a small, tetraploid perennial herb. It grows in rock fissures on mountain slopes, gravel mounds, and alpine meadows at elevations of 4800-5300 m in Xizang (Tibet). In addition to pollen traits, it shares many other morphological features with P. variifolium, as indicated by the results of our field observations. For instance, their disciform capitula have urceolate involucres consisting of 4 rows of involucral bracts with dark brown margins. The disc florets have brownish style-branches and erect corolla lobes. Taking both the molecular [ Figure 2] and morphological data into consideration, we regard the tetraploid A. purpurea as a member of Phaeostigma, which has been adapted to the alpine environment probably via polyploidization.
With respect to other three genera, Elachanthemum, Kaschgaria, and Brachanthemum, cpDNA data show them to be quite divergent ( Figure 2A), but nuclear CDS exon data place them in a clade sister to the Chrysanthemum+Ajania+Phaeostigma group ( Figure 2B). Given the pattern of molecular and morphological variations, particularly pollen morphology, the relationships among these genera are in agreement with Tzvelev's hypothesis [53], i.e., the radiateheterogamous and disciform-heterogamous taxa may have evolved from the same 'dendranthemoid' ancestor.

Main lines of phylogeny of Chrysanthemum
Both the cpDNA and nuclear CDS gene data suggest that C. sinuatum, C. arisanense, and C. naktongense (ZJ population) are distantly related to the main lineage of Chrysanthemum sensu stricto ( = Chrysanthemum+Ajania) (Figures 2-4). C. sinuatum, and C. arisanense are diploid species (Table 1 & reference [12]), and C. naktongense has various ploidy levels (2x to 4x and 6x) ( Table 1). The population ZJ of C. naktongense that is clustered with C. arisanense and close to C. sinuatum is diploid. These three species are morphologically distinct and geographically separated. C. sinuatum is a narrowly distributed species only in Altai, Russia [12]. It has pink ray florets and leaves with 2-5 pinnatipartite or pinnatisect. While, C. arisanense is endemic to Taiwan and has yellow ray florets and leaves with 1-2 pinnatipartite [8]. C. naktongense is quite different from the former two. It is polymorphic not only in ploidy but also in morphology, and therefore, there was much controversy over its species identity [8,9,63]. The present data also show marked molecular differences for two accessions of C. naktongense (ZJ and NML). In our opinion, this is not a good species, and it needs multidisciplinary data from multiple populations to resolve this species problem.
Chrysanthemum has the species diversity center in China. According to the geographic distribution and morphology, the Chinese Chrysanthemum species can be divided into two species groups, the C. indicum group (clade III in Figure 3 and mainly clade I in Figure 4) and the C. zawadskii group (clade IVa in Figure 3 and mainly clade II in Figure 4). The C. indicum group is distributed in southern China. It is mainly represented by C. indicum with creeping stems and small capitula with yellow ray florets. In contrast, the C. zawadskii group occurs in northern China, predominated by C. zawadskii with erect stems and large capitula with white-purple ray florets. At the diploid level, C. lavandulifolium and C. nankingense belong to the C. indicum group, and C. mongolicum and C. chanetii are of the C. zawadskii group.
In the CDS gene tree (Figure 4), more geographic patterns can be recognized: Clade III is composed of narrowly distributed Chrysanthemum species in alpine or isolated mountains except two Ajania species. Within clade IV, molecular variation shows geographic structure. For example, the group IVa contains species/varieties mostly from Japan and Korea except three haplotypes of the hexaploid C. vestitum, and IVb only consists of those from China and Japan. Two hexaploid chrysanthemum cultivars, C. morifolium cv. HSZ and C. morifolium cv. ZY, are both Figure 4. Bayesian 50% majority consensus tree of Chrysanthemum, Ajania, and Phaeostigma. The tree is based on the nuclear CDS exon+intron sequences. Posterior probabilities from the Bayesian inference (PP)/bootstrap percentages (BP) from the maximum parsimony analysis for the major clades are shown next to the nodes. ''-'' indicates where the BP value is ,50% and the corresponding clade collapsed in the MP 50% bootstrap majority consensus tree. Labels of terminal branches contain: species or variety name followed by population code (when available; in capital letters), and then clone identities (from ''a'' to ''e''). doi:10.1371/journal.pone.0048970.g004 Despite some incongruence, data from the plastid and nuclear DNA all show the geographic structure of the infrageneric diversification within Chrysanthemum. There are also many endemic species in Chrysanthemum, and most of them, such as C. hypargyrum, C. rhombifolium, C. yoshinaganthum, and C. yezoense, only occur in a specific habitat [1,4,8,9]. All these findings suggest that the geographic and ecological speciation plays an important role in the evolution of the genus Chrysanthemum.

Hybridization and polyploidy in the evolution of Chrysanthemum
Previous studies have demonstrated that polyploidy and hybridization are common in Chrysanthemum [1][2][3][4][5]. Both the nuclear and the cpDNA data presented here give some clues to the participation of various diploids in the origins of the polyploid taxa. For example, the diploids C. lavandulifolium and C. nankingense may have been involved in the origin of the polyploids of the C. indicum group (Figures 3 and 4). This conclusion is supported by previous studies [2,24,25,64,65]. Similarly, the diploids C. mongolicum and C. chanetti may have been involved in the origin of the polyploids of the C. zawadskii group (Figures 3 and 4). Within each group, species are very similar morphologically, but the diploid can be distinguished from those of polyploids by their leaf shape and relatively small capitula [2,5].
Polyploidy via interspecific hybridization (allopolyploidy) is a major mechanism of speciation in plants [66,67]. This mechanism can be involved in the speciation of some polyploid species in Chrysanthemum, e.g., C. argyrophyllum. This is a perennial hexaploid species occurring in the rock crevices of mountain peaks in NW Henan and SE Shaanxi Provinces. In the cpDNA tree (Figure 3), the haplotype of C. argyrophyllum is resided in the clade IVa that includes the diploid C. mongolicum. In the nuclear CDS gene tree (Figure 4), three haplotypes of C. argyrophyllum are clustered with the tetraploid C. indicum (SNJ) in clade I, and another two haplotypes are nested in clade II close to the diploid C. mongolicum.
Morphologically, C. argyrophyllum is more close to C. mongolicum than C. indicum by sharing more common morphological characters, e.g., leaf-like involucral bracts and white-purple ray florets [8]. Thus, all the evidence from cpDNA, nuclear CDS and morphology support that C. argyrophyllum is an hexaploid species with hybrid origin, and diploid C. mongolicum may be its maternal progenitor.
The current study produced little evidence of autopolyploidy. This is consistent with earlier studies [68]. The 4x population GZ of C. indicum may be autopolyploid, because C. indicum (GZ) is nested in the clade III of cpDNA tree including the diploid C. lavandulifolium and the CDS haplotypes of C. indicum (GZ) only form a clade with that of C. lavandulifolium. In addition, the data from RAPD and ISSR markers showed that only the C. lavandulifoliumspecific bands in those of three putative diploid progenitors can be found in the tetraploid C. indicum (GZ) [2]. All evidence was consistent with the possibility that C. indicum (GZ) might have an autopolyploid origin.
Considering all the evidence from the present molecular data and previous studies [1,2], we conclude that geographic and ecological conditions as well as hybridization and polyploidy have played important roles in the divergence and speciation of the genus Chrysanthemum.