Tao Sang, Daniel J. Crawford &Tod F. Stuessy

Chloroplast DNA Phylogeny, Reticulate Evolution, and Biogeography of Paeonia (Paeoniaceae)

American Journal of Botany 84(S): 1120-1136. 1997.

Department of Plant Biology, Ohio State University, Columbus, Ohio 43210

The coding region of the matK gene and to intergenic spacers, psbA-trnH and trnL(UAA)-trF(GAA), of cpDNA were sequenced to study phylogenetic relationships of 32 Paeonia species. In the psbA-trnH intergenic spacer, short sequences bordered by long inverted repeats have undergone inversions that are often homoplasious mutations. Insertions/deletions found in the two intergenic spacers, mostly resulting from slipped-strand mispairing, provided relatively reliable phylogenetic information. The matK coding region, evolving more rapidly than trnL-trnF spacer and more slowly than the psbA-trnH spacer, produced the best resolved phylogenetic tree. The matK phylogeny was compared with the phylogeny obtained from sequences of internal transcribed spacers (ITS) of nuclear, ribosomal DNA. A refined hypothesis of species phylogeny of section Paeonia was proposed by considering the discordance between the nuclear and cpDNA phylogenies to be results of hybrid speciation followed by inheritance of cpDNA of one parent and fixation of ITS sequences of another plant. The Eurasian and western North American disjunct distribution of the genus may have resulted from interruption of the continuous distribution of ancestrial population of extant peony species across the Bering land bridge during the Miocene. Pleistocene glaciation may have played an important role in triggering extensive reticulate evolution within section Paeonia and shifting distributional ranges of both parental and hybrid species.

Key words: biogeography; chloroplast DNA; hybridization, matK; Paeonia; psbA-trnH spacer; reticulate evolution; trnL-trnF spacer.

Paeonia comprises -35 species of shrubs and perennial herbs distributed widely in five disjunct areas in the northern hemisphere: eastern Asia, central Asia, the Western Himalayas, the Mediterranean region, and Pacific North America (Stern, 1946; Pan, 1979; Tzanoudakis, 1983: Pei, 1993). The genus is systematically isolated, having been placed in the unigeneric family Paeoniaceae, which has either been placed by itself or together with Glaucidiaceae in order Paeoniales (Takhtajan. 1969, 1987; Thorne, 1992). Because of their great ornamental and medicinal value, peonies have been known as "king of flowers" in China and "queen of herbs" in Greece for > 1000 yr (Ganibrill, 1988).

Three sections are recognized within Paeonia (Stern, 1946). Section Oneapia, endemic to Pacific North America, comprises two herbaceous species with conspicuous staminodial disks and small fleshy concave petals. Section Moutan with six species, occuring in central and western China, was divided into two subsections, Delavayanae and Vaginatae. They are shrubs with conspicuous staminodial disks and large spreading petals. Section .Paeonia ("Paeon"), which includes the type species P. officinalis, was also divided into two subsections, Foliolatae and Paeonia ("Dissectifoliae"), distributed disjunctly in eastern Asia, central Asia, the western Himalayas, and the Mediterranean region. This section consists of -27 herbaceous species with inconspicuous or no staminodial disks and large petals that are either spreading or cup-shaped. Sections Oneapia and Moutan contain only diploid species (2n = 10), while one-third of the species in section Paeonia are tetraploids (Stern, 1946; Tzanoudakis, 1983; Hong, Zhang, and Zhu, 1988).

Paeonia is a phylogenetically and taxonomically complex group (Stebbins, 1938a; Hong, Zhang, and Zhu, 1988). In particular, section Paeonia may have undergone complex reticulate evolution that further obscured phylogenetic relationships (Stebbins, 1948; Sang et al., 1995). Regarding origins of tetraploid species in this section, Barber (1941) and Stern (1946) considered them autotetraploids derived from certain extant diploid ancestors. In contrast, Stebbins (1948) argued that the majority of tetraploid species are allotetraploids, based on observations of bivalents in meiosis of some tetraploid species, such as P. officinalis, P. peregrina, and P. wittmanniana. He further indicated that certain tetraploid species appeared to link gaps of morphological Variation among some diploid species, which also suggested hybrid origins of the tetraploids. Later cytogenetic studies confirrned allotetraploid origins of P. officinalis and P. peregrina, and also revealed P. parnassica as an allotetraploid (Tzanoudakis, 1983; Schwarzacher-Robinson, 1986). Recent phylogenetic studies of Paeonia using sequences of internal transcribed spacers (ITS) of nuclear ribosomal DNA supported Stebbins' hypothesis and documented reticulate evolution in section Paeonia (Sang, Crawford, and Stuessy, 1995). The hybrid species were identified by full or partial ITS sequence additivity of the putative parental species. Partial sequence additivity was suggested to be a result of parental homogenization of parental ITS sequences via gradients of gene conversion (Sang, Crawford, and Stuessy, 1995). Observation of partial homogenization of parental ITS sequence additivity in hybrid species implies the possibility of complete homogenization of parental sequences in certain hybrid species whose hybrid origin, thus, could not be detected by ITS sequences (Wende, Schnabel, and Seelanan, 1995). Also, diploid hybrid species may have lost one of the parental ITS sequences through segregation. One way to test this hypothesis is to compare the ITS phylogeny with the cpDNA phylogeny of the section. lf a hybrid species fixes ITS sequences from one parent via gene conversion, but inherited cpDNA from the other patent, it will have discordant positions on the ITS and cpDNA phylogenies (Rieseberg and Soltis, 1991; Soltis and Kuzoff, 1995; Wolfe and Elisens, 1995).

The coding region of the matK gene, and two intergenic spacers of cpDNA, trnL-trnF and psbA-trnH, were sequenced for phylogenetic reconstructions. The matK gene, encoding a maturase, has been suggested as the most rapidly evolving coding region found so far in the chloroplast genome (Neuhaus and Link, 1987; Olmstead and Palmer, 1994). Sequences of the matK coding region have been used to assess phylogenetic relationships at the intrafamilial level in angiosperms, such as within Polemoniaceae (Steele and Vilgalys. 1994) and Saxifragaceae (Johnson and Soltis, 1994). Noncoding regions of cpDNA , which are presumably under less functional constraint and thus evolve more rapidly, may also provide useful phylogenetic information at the lower taxonomical levels (Clegg et al., 1994: Gielly and Taberlet, 1994). A few noncoding regions, including introns and intergenic spacers, have been sequenced recently to assess intrafamilial and intrageneric relationships.(Taberlet et al., 1991, Golenberg et al., 1993: Morton and Clegg, 1993; Bohle et al., 1994; Ham et al., 1994; Mauer, Natali, and Ehrendorfer. 1994; Mes and Hart, 1994). The trnL-trnF intergenic spacer represents the most frequently used noncoding region of cpDNA in phylogenetic studies (Bohle et al., 1994; Gielly and Taberlet, 1994; Ham et al., 1994; Mes and Hart, 1994). The psbA-trnH intergenic spacer, an evolutionarily plastic region (Aldrich et al., 1988), is employed as a new phylogenetic marker to assess interspecific relationships in Paeonia.

Paeonia, with widely disjunct distributions and rich endemismn, provides a favorable system for studying historical biogeography of the northern hemisphere. A marked difficulty in understanding historical biogeography of the northern hemisphere is that significant shifts of distributional ranges and extinction of taxa may have been caused by Pleistocene glaciation (Noonan, 1988; Potts and Behrensmeyer, 1992). Pleistocene glaciation was suggested as a primary factor triggering extensive hybridization in section Paeonia and subsequently changing its distributions (Stebbins, 1948). Reconstruction of reticulate evolution in the section using ITS sequences transcribed spacers (ITS) of nuclear ribosomal DNA supported Stebbins' hypothesis and documented reticulate evolution in section Paeonia (Sang, Crawford, and Stuessy, 1995).

The primary purposes of this paper, therefore, were to:

  1. reconstruct cpDNA phylogeny of Paeonia, and assess molecular evolution and phylogenetic utility of coding and noncoding regions of cpDNA;

  2. compare the cpDNA phylogeny with the ITS phylogeny of the genus, and particularly of the section Paeonia in order to gain new insights into complex reticulate evolution within the section; and

  3. discuss biogeography of Paeonia based on molecular data and phylogenetic information.


Thirty-seven accessions of 32 Paeonia species were sequenced. For most species, fresh leaves used as sources of DNA were collected from natural population, in Bulgaria, China, Greece and Spain. The voucher specimens are deposited in OS and UPA. The remaining species were collected from the Royal Botanic. Gardens, Kew, and the Beijing Botanical Garden (Table 1). The limited intraspecific sampling is due partly to the remote and endemic distributions, and rarity of most peony species. Particularly in section Paeonia. about one-third of species are endemic to a single island, mountain range, or other small area (Stern, 1946; Pan. 1979; Tzanoudakis. 1983). Total DNA was isolated from leaf tissues using the CTAB method (Doyle and Doyle, 1987), and purified in CsCl/lethidium bromide gradients. Double-stranded DNAs were amplified by 30 cycles of symmetric PCR (Sang et al., 1995). The amplification products were purified by electrophoresis through 1.0% agarose gel followed by use of Bioclean, (U. S. Biochemical). Purified double-stranded DNAs were used for sequencing reactions employing Sequenase Version 2.0 (Amersham CO., Arlington Heights, IL) (Sang et al., 1995). The sequencing reaction products were separated electrophoretically in 6% acrylamide gel with wedge spacers for 3 h at 1500 V. After fixation, gels were dried and exposed to Kodak XAR x-ray film for 18-48 h. DNA sequences were aligned manuallv. Sequences of the matK coding region were aligned with those of tobacco and mustard (Neuhaus and Link. 1987)

Primers for amplifying and sequencing the matK coding region and two intergenic spacers are given in Table 2.

Table 2::

The forward PCR primer matK 1 F, is located -70 base pairs (bp) upstream of the start codon of the matK coding region and is sequences are conserved between tobacco (Shinozaki et al.. 1986) and mustard (Neuhaus and Link, 1987). The reverse PCR primer, matK1R, is similar to the primer, trnkR. of Steele and Vilñgalys (1994) except for one more nucleotide at its 5' end. The four internal matK primers are also located in the conserved regions. These six primes have bee used successfully to amplify and sequence the matK coding region in Aesculus of Hippocastanaceae (except matKR2) (Q.Xiang and D. J. Crawford. unpublished data). Coreopsis of Asteraceae (S.-C. Kim and E). 3, Crawford, unpublished data), and Hamamelidaceae (except matKF-l) (J. Li. unpublished data. University of New Hampshire, Durham, NH).

The forward primers (psbAF) for amplifying the psbA-trnH intergenic spacer, were designed in a region of the psbA genes conserved between tobacco of Solanaceae (Shinozaki et al., 1986) and Brassica napus of Brassicacea (Genbank No. M36720). The reverse primer (trnHR) is in a region of the trnH gene that is conserved among tobacco, Helianthus annus of Asteraceae (Genbank No. X604218), and Arabidopsis thaliana of Brassicaceae (Genbank No. X79898), The primers for the trnL-trnF intergenic spacer are similar to those of Taberlet et al. (1991). The reverse primer, trnFR, has one more nucleotide at the 5' end of the primer f of Taberlet et l. (1991). The forward primer (trnLF) is designed ten nucleotides farther away from the 3' end of the trnL 3' exon than the primer of Taberlet et al. (1991) in order to read sequences closer to the 5' end of the intergenic spacer.

Sequence divergence between species for each cpDNA region was calculated using DNADIST program of PHYLIP version 3.5c (Felsenstein, 1994). The Jukes-Cantor model was used for correcting possible multiple hits of nucleotide substitutions (Jukes and Cantor, 1969).

Phylogenetic analysis were performed initially using the matK coding region and the psbA-trnH intergenic spacer, independently, and then using the combined data set from mutations in the three cpDNA regions. Mutations, including nucleotide substitutions and insertions/deletions (indels), were analyzed by unweighted Wagner parsimony using PAUP version 3. 1.1 (Stafford, 1993). Indels were coded as binary characters in the analysis. The shortest trees were searched with TRR Branch Swapping of the heuristic method, and character changes were interpreted with the ACCTRAN optimization. Bootstrap analysis were combined out with 500 applications using TBR Branch Swapping of the Heuristic search (Felsenstein, 1985). Section 0neapia of Paeonia was chosen as the outgroup for the cladistic analysis of the genus (Watrous and White, 1981; Maddison, Donoghue, and Maddison, 1984; Sang, Crawford and Stuessy, 1995).

Discordance between the ITS and matK phylogenies was assessed by two methods, inspection and assessment of support, and the Templeton test (Mason-Gamer and Kellogg, in press). These two methods are chosen because they can assess conflicts involving individual clades between the trees. The method of inspection and assessment of support compares bootstrap support for a certain clade in two data sets. For example, when a clade is strongly supported by a bootstrap value of 90% on tree A, but does not appear on tree B, we an check the table of "partitions found in one or more trees and frequency of occurrence” from the PAUP 3. 1.1 bootstrap output of the B tree for any possibility support for this clade. The lower the support the higher is the conflict for formation of this clade between the two data sets (Mason-Gamer and Kellogg, in press). The Templeton test forces data set A to generate the tree topology obtained from data set B, and tests whether the resulting tree is significantly less parsimonious than the most parsimonious tree obtained from data set A under no constraint (Templeton, 1983; Larsen, 1994). When tree B contains only one resolved clade that is used as the constraint topology for parsimony analysis of data set A, the significant level of support of this clade by data set A is tested (Mason-Gamer and Kellogg. in press). Because only ITS sequences that did not show additivity were used in parsimony analysis (Sang, Crawford, and Stuessy. 1995), the matK data set is reduced to match the same species in the ITS data set for comparison.


The entire matK coding region is 1491 bp long, and nucleotide substitutions were found at 53 nucleotide sites among all the peony species. No nucleotide substitutions were found between sequences of different accessions of the same species. Parsimony analyses generated five equally most parsimonious trees with length of 59, a consistency index (CI) of 0.949, and a retention index (RI) of 0.972. A strict consensus tree of These five trees was computed and has a tree length of 62, a CI of 0.903 (0.842 excluding autapomorphies), and a RI of 0.945 (Fig. 1).

Aligned sequences of the psbA-trnH intergenic spacer of Paeonia species are shown in Fig. 2. The sequences of two populations of P. lutea differ by one nucleotide substitution. For the remaining species, different accessions of each species have identical sequences. With this straightforward sequence alignment, the length of the intergenic spacer varies from 281 bp (P. mairei) to 324 bp (P. spontanea and P. szechanica). A total of 31 variable sites (with nucleotide substitutions) and 13 indels occur among These species. Twenty-four variable sites and 11 indels were used for reconstructing phylogeny and nine equally most parsimonious trees, with a length of 36, CI of 0.946, and RI of 0.98 1, were obtained (sec Discussion for reasons for excluding some mutations from the phylogenetic analysis). A strict consensus tree with a length of 37, CI of 0.921 (0.875 excluding autapomorphies), and RI of 0.971, was generated (Fig. 3).

For the trnL-trnF intergenic spacer, because the forward primer is still very close to the 3' end of the trnL 3' exon about ten nucleotides at the 5' end of the spacer could not be read. Sequences of different accessions of the same species am identical. Length of the aligned sequences varies from 372 bp (P. obovata) to 404 bp (P. sterniana). Among all species, only nine variable sites and five indels were detected (Table 3). Phylogenetic reconstruction was not performed for this region alone because of little phylogenetic information.

Phylogenetic analysis of combined mutations from the matK coding region and the two intergenic spacers generated 102 equally most parsimonious trees. The consensus tree, with a CI of 0.855 (0.769 excluding autapomorphies) and a RI of 0.926, differs topologically from the matK phylogeny by only two clades in section Paeonia (tree not shown). The clade containing P. veitchii, P. emodi, and P. xinjiangensis on the matK phylogeny collapsed on the combined tree, and a new clade containing P. clusii and P. mascula formed on the combined tree.

Comparisons of sequence divergence and phylogenetic information from variable sites among the two intergenic spacer and the mark coding region are given in Table 4. Comparisons of phylogenetic information from indels and relative frequency of indels vs. nucleotide substitutions between the two intergenic spacers are given in Table 5. Average percentage sequence divergences of matK, the psbA-trnH intergenic spacer, and ITS were compared with and among sections of Paeonia (Table 6). Comparisons of sequence divergences indicate that ITS sequences evolve slightly more rapidly than the psbA-trnH intergenic spacer, and over three times more rapidly than the matK coding region.

Comparisons of the phylogenetic tree generated from the ITS data set and the reduced matK data set are illustrated in Fig. 4. The bootstrap values from the two data sets were compared for all the clades, and suggest highly conflicting support for several clades of section Paeonia an either ITS (Fig. 4I, II). The two well-supported basal clades of section Paeonia on the ITS tree received no support from the matK sequences. The strongly supported sister group relationship of P. veitchii and P. xinjiangensis an the ITS tree receives no support from the mark data. the monophyly of P. tenufolia together with P. arietina, P. humilis, P. officinalis, and P. parnassica was supported by 100% bootstrap an the ITS tree, but 0% by the matK data. On the matK phylogeny, a well-supported monophyletic group (P. obovata, P. japonica, P. arietina, P. humilis, P. officinalis, and P. parnassica) had only 0.3% bootstrap support from the ITS data set. A group of four species (P. arietina, P. humilis, P. officinalis, and P. parnassica), which have identical ITS and matK sequences, was supported by 65% bootstrap value an the matK tree, but an the ITS phylogeny they never formed a monophyletic group by themselves without P. tenuifolia. likewise, the moderately supported (63%) sister group relationship between P. lactiflora and P. xinjiangensis an the matK tree was not supposed at all by the ITS data without the involvement of P. veitchii. .

For the Templeton test, two basal clades of section Paeonia and a clade consisting of P. lactiflora and P. xinjiangensis, which are found only an die ITS tree, were used as constraint topology for generating parsimonious trees from the reduced matK data set (Fig. 4I). Similarly, each of three clades found only an the matK tree was used as constraint topology for the ITS data set (Fig. 4II). In none of the cases, however, was the exact constraint topology obtained from parsimony analysis (Fig. 4A-E). Examination of the character of each data set indicates that not a single character in a data set can serve as a synapomorphy for a clade obtained only from the other data set. This result supports the comparative bootstrap values in demonstrating that the two data sets highly conflict in supporting each of these clades. In the first test (Fig. 4A), it is significantly less parsimonious (P < 0.005, two-third value) to form a clade including P. japonica, P. obovata, P. humilis, P. officinalis, P. arietina, and P. tenuifolia with the ITS data set, although the cost of taking out P. tenuifolia is still impossible to erasure. The results of the remaining four tests, however, are not significant because the number of characters undergoing step changes during each constraint analysis is smaller than the minimal requirement of five characters in order to get significant test results (Fig. 4B-E). Such results may be due to the failure of obtaining the constraint topology after parsimony analysis. In each of the eases, only the cost of breaking down a certain clade is taken into account while formation of a desired clade was not achieved, and thus the cost could not be estimated.


Inversions in the psbA-trnH intergenic spacer

The simple alignment of sequences between nucleotide sites 57 and 97 of the psbA-trnH intergenic spacer (Fig. 2) may not be an appropriate evolutionary interpretation. This region contains a pair of 20 or 27 bp exact inverted repeats, which are separated by 21 or 6 bases, respectively (Fig. 5A). Three types of sequences can be recognized: type I occurs in section Oneapia, and P. suffruticosa ssp. spontanea and P. szechanica of subsection Vaginatae; type II occurs in subsection Delavayanae and P. rockii; and type III is found in section Paeonia, which further includes two subtypes i and ii (Fig. 5A). Type I and II sequences can be converted into each other by inversions of sequences bordered by the inverted repeats. Similarly, in type III sequence, an inversion of the sequence bordered by the inverted repeats in one subtype sequence can give rise to the other subtype. The inversions occur quite frequently and homoplasiously within the genus (Fig. 2). In contrast to some large inversions in cpDNA that provided reliable phylogenetic information at the higher taxonomic levels (Jansen and Palmer, 1987; Doyle et al., 1992; Rauberson and Jansen, 1992), short inversions in the intergenic spacer easily yield homoplasious information even at the interspecific level and thus should not be included in phylogenetic analyses.

The mechanism responsible for change between the type III sequence and the other two types is more complex. In the type I sequence, there is another pair of short inverted repeats in the region between the long inverted repeats (Fig. 5B). Therefore, a stem-loop structure with two stems and two loops can be formed (Fig. 5B). The evolutionary changes that are likely to have occurred in the small loop between the two stems include deletion of the T and two transitions of A to G to match the two Cs so that a single longer stem of the type 111 sequence could be formed. The two substitutions in this small loop are facilitated by this particular stern-loop structure, and thus

should not be treated as regular substitutions in calculating sequence divergence in order to avoid overestimating rates of sequence divergence. Therefore, they were not taken into account in calculating sequence divergence or reconstructing phylogeny. Likewise, the two substitutions, A to C at site 76 and T to G at site 97, are at the corresponding positions of the inverted repeats, and treated as only one substitution for sequence divergence estimation and phylogenetic reconstruction.

Insertions/deletions in the intergenic spacers

Of 11 indels in the psbA-trnH intergenic spacer, four are perfect (indels 8 and 11) or imperfect (indels 2 and 3) duplications or deletions of prior duplications of adjacent sequences (Fig. 2). Slipped-strand mispairing is most likely the mechanism responsible for this type of indel (Levinson and Gutman, 1987). Indels 7, 9, and 10, which are duplications or deletions of a portion of poly(T) tracks, may also result from slipped-strand mispairing (Wolfson, Higgins, and Sears, 1991). Since the probability of occurrence of further insertions or deletions increases as the track of repetitive nucleotide sequences gets longer (Streisinger and Owen, 1985; Golenberg et al., 1993), multiple indels must have occurred at indels 7 and 9 to create the pattern of differential lengths of poly(T) tracks. In the trnL-trnF intergenic spacer, three of six indels are also portions of poly(T) tracks.

Apparently, indels in these two intergenic spacers occur less frequently and provide a smaller amount of phylogenetic information than nucleotide substitutions (Table 5). As phylogenetic characters, indels do not conflict with each other or with nucleotide substitutions in either of the two intergenic spacers. Only indel 8 of the psbA-trnH intergenic spacer conflicts with relationships in the matK phylogenies (Figs. 1, 2). Overall, the indels in the cpDNA intergenic spacers are quite reliable phylogenetic characters in Paeonia, which is concordant with findings in other plant groups at the lower taxonomical levels (Ham et al., 1994; Mes and Hart, 1994). However, indels may not be reliable phylogenetic characters at higher taxonomic levels because the chance of superimposition of indels increases as divergence time increases (Morton and Clegg, 1993; Golenberg et al., 1993).

Nucleotide substitutions in cpDNA coding and noncoding regions

A comparison of average species pairwise sequence divergence in the two intergenic spacers and the matK coding region (Table 4) indicates that the psbA-trnH intergenic spacer has much higher rates of nucleotide substitutions than the other two regions. The trnL-trnF intergenic spacer has lower substitution rates than the matK coding region, suggesting that higher substitution rates should not always be expected in noncoding regions than in coding regions of cpDNA. Distinguishing autapomorphic, synapomorphic, and homoplasious substitutions in the intergenic spacers and the matK coding region should enable comparisons of the quality of phylogenetic information yielded from these regions (Table 4). The percentage of phylogenetically informative sites (the sites where substitutions are shared by two or more taxa) among the variable sites is similar for the three regions. The percentage of synapomorphic sites among informative sites is highest in the psbA-trnH intergenic spacer (92.3%), and lowest in the trnL-trnF intergenic spacer (66.7%). Therefore, the psbA-trnH spacer, which evolves most rapidly among the three regions and provides best synapomorphic information, should be a useful region for phylogenetic studies at the lower taxonomical levels. The matK coding region, although evolving about twice as slowly as the psbA-trnH spacer, has over twice as many synapomorphic sites as the intergenic spacer because it is about four times longer. In comparison with the ITS phylogeny, the matK coding region provides a comparable amount of information for phylogenetic reconstruction, e.g., a similar number of substitutions is found to support each section of the genus (Figs. 1, 6), and may serve as a good marker for phylogenetic studies at the intrageneric level. The most frequently used intergenic spacer, trnL-trnF, however, evolves most slowly and homoplasiously, and thus its phylogenetic utility at the intrageneric level is questionable.

Phylogenetic reconstruction

phylogenies reconstructed from sequences of the mark coding region and psbA-trnH intergenic spacer are resolved and congruent at the sectional level (Figs. 1, 3). Two subsections of section Moutan are also recognized in the cpDNA phylogenies. Lower resolution of relationships within section Paeonia is probably due to the relatively recent origins of the species and reticulate evolution that led to the loss of one divergent cpDNA (sec later discussion). The CL (0.903) and RI (0.945) of the matK consensus tree are high, suggesting that the phylogenetic reconstruction is fairly reliable. In fact, only three characters (variable sites) out of 53 characters are homoplasious for the parsimony analysis. The phylogenetic analysis of the psbA-trnH intergenic spacer resolved only two clades within section Paeonia, The clade containing 22 species support--d only by indels 8 conflicts with relationships on the matK phylogeny (Figs. 1, 2).It is very unlikely that this clade reflects the true cpDNA phylogeny, but is just a result of random deletions of the AIT duplication in P. emodi, P. sterniana, P. clusii, and two subspecies of P. mascula. This homoplasious mutation is also responsible for the collapse of the matK clade containing P. veitchii, P. emodi, and P. sterniana on the tree generated from the combined dataset.

The trnL-trnF intergenic spacer provides very limited phylogenetic information (Table 3). Two indels distinguish section Oneapia from the other two sections. The indel and one nucleotide substitution serve as synapomorphies for section Paeonia. One substitution defines subsection Vaginatae. One substitution suppose the sister relationship of P. lactiflora and P. xinjiangensis, as on the psbA-trnH spacer and matK phylogenies. One shared substitution by P. californica and P. rockii, however, is clearly homoplasious. Another apparent homoplasious substitution shared by eight species (OBO, STE, WIT, BAN, ARI, HUM, OFF, and PAR) may result from a synapomorphic substitution that defines the monophyletic group on the matK phylogeny (JAP, OBO, WIT, BAN, ARI, HUM, OFF and PAR)'followed by a reversal substitution in P.japonica. This explanation is in agreement with the matK and ITS phylogenies where P. japonica is a sister group of F. obovata. The reason why P. sterniana has this substitution is unclear.

Overall, the matK coding region served as a better phylogenetic marker for resolving close specific relationships in Paeonia than the intergenic spacers. The matK phylogeny will be compared with the ITS phylogeny for a better understanding of the species phylogeny of the genus.

DNA sequence and morphological divergence

Evolutionary tempos may or may not be concordant at molecular and morphological levels (Sytsma and Smith, 1992). Among three sections of Paeonia, section Oneapia is the most distinct one at the molecular level. A comparison of sequence divergence between any one section and the other two indicates that section Oneapia has the highest percentage sequence divergence values of ITS (4.38), psbA-trnH intergenic spacer (3.74), and matK (1.33) (Table 6).The earliest evolutionary split within the genus Paeonia might have occurred between section Oneapia and the other two sections if the molecular clock is assumed. Morphologically, section Oneapia is also distinct from the other two sections by its small flowers (2-3 cm in diameter vs. > 5 cm in sections Moutan and Paeonia) with fleshy and strongly concave petals.

Within section Oneapia, however, rates of divergence between DNA sequences and morphology are strikingly different. Sequence divergence is higher within-section Oneapia than within any other two sections (Table 6). The two species of section Oneapia are distributed allopatrically, i.e., P. californica is endemic to southern California, and P. brownii is found from northern California to British Columbia. Paeonia californica is adapted to warm and wetter climates and flowers from February to April, where P. brownii is semixerophytic and flowers during June and July (Stebbins, 1938b). However, these two species are morphologically very similar to each other and had been treated as one species until detailed morphological, ecological, and cytological studies that suggested they have undergone considerable genetic divergence rather than only morphological modification due to ecological factors (Stebbins, 1938b; Stebbins and Ellerton, 1939). DNA sequence data support this hypothesis and indicate that morphological evolution in section Oneapia has been remarkably slow compared with the high level of sequence divergence.

Within subsection Vaginatae of section Moutan, however, morphological divergence apparently exceeds DNA sequence divergence. Three species studied by DNA sequences are morphologically distinct arid allopatrically distributed, but have identical ITS sequences. In addition, P. suffruticosa ssp. spontanea and P. szechanica have identical sequences of psbA-trnH intergenic spacer, and P. rockii and P. szechanica have identical matK sequences.

Reticulate evolution in section Paeonia

Phylogenies of section Paeonia obtained from ITS and matK sequences are concordant in certain respects and discordant in others (Figs. 1, 6). Both phylogenies support monophyly of each of the three sections of Paeonia as well, as subsections of section Moutan. They differ substantially within section Paeonia (Fig. 4) where reticulate evolution has been documented previously by morphological, cytogenetic, and molecular data (Stebbins, 1948; Tzanoudakis, 1983; Sang, Crawford, and Stuessy, 1995). Discordance between the ITS and matK phylogenies, therefore, may result from hybrid speciation followed by inheritance of cpDNA from one patent and fixation of the ITS sequences from another parent (Rieseberg and Soltis, 1991; Soltis and Kuzoff, 1995). Since maternal transmission of cpDNA has been found in the majority of flowering plants, the patent whose cpDNA is transmitted to hybrids is very likely the maternal parent (Corriveau and Coleman, 1988-, Hanis and Ingram, 1991; Mogensen, 1996). A synthesis of both gene phylogenies leads to a more refined hypothesis of species phylogeny reflecting both divergent and reticulate evolution (Fig. 7). Besides hybrid speciation suggested directly by ITS sequence ad

for matK) (Figs. 1, 6). The ancestral populations of the lineage represented by the smaller ITS clade which might not have been involved in the hybridization, may have gone extinct. Therefore, two early divergent types of ITS sequences are maintained but one of the early divergent types of chloroplast genomes was lost after the hybridization.

The same explanation can be given to account for the hybrid origin of species P. xinjiangensis, P. japonica, P. obovata, P. wittmanniana, P. arietina, P. humilis, P. officinalis, and P. parnassica, which also have discordant positions in the two gene phylogenies (Figs. 1, 6, 7). Paeonia xinjiangensis forms a strongly supported sister group with P. veitchii on the ITS phylogeny (98% bootstrap value), but switches its sister group relationship to P. lactiflora on the matK phylogeny, suggesting that P. xinjiangensis is a hybrid that fixed ITS sequences of P, veitchii and inherited cpDNA from P. lactiflora. Likewise, P. japonica and P. obovata, which are distantly separated from four species (P. arietina, P. humilis, P. officinalis, and P. parnassica) on the ITS phylogeny, become the sister groups to them on the matK phylogeny, indicating that P. japonica and P. obovata were derived from hybridization between the lineage containing these four species as the cpDNA donor and the other parent with the type of ITS sequences belonging to the smaller clade of the ITS phylogeny. By the same reasoning, the hybrid origin of P. wittmanniana is also suggested (Figs. 1, 6, 7).

Four species P. arietina, P. humilis, P. officinalis, and ,P. parnassica., and P. tenuifolia form a strongly supported clade (100% bootstrap value) on the ITS phylogeny, but become two separate lineages on the matK phylogeny. The suggest that the four species were derived through hybridization with P. tenuifolia serving as one patent. The other parent was either from an extinct basal lineage on the matK phylogeny or it still cannot be identified by the present data. The hybrid species subsequently fixed ITS sequences of P. tenuifolia and thus becomes its sister group on the ITS phylogeny, but exists as an independent clade on the matK phylogeny as did its cpDNA parent. This hypothesis is in agreement with the previous cytogenetic studies that revealed that P. arietina, P. officinalis, and P. parnassica are allotetraploids (Stebbins, 1948; Tzanoudakis, 1983; Schwarzacher-Robinson, 1986).

lf a hybrid species inherited cpDNA and fixed ITS sequences from the same parents however, the hybridization would not be detected by such a comparison. The present reconstruction, therefore, may still be an underestimate of reticulate evolution in Paeonia. Solutions to this problem include increase of intraspecific sample size (Doebley, 1989) and reconstruction of independent nuclear gene phylogenies. Nonetheless, the present reconstruction (Fig. 7) should represent more closely the species phylogeny than either ITS or matK Phylogeny alone, and thus serve as a hypothesis to be tested with additional data.

An alternative explanation of discordance between nuclear and cpDNA phylogenies is lineage sorting (Doyle, 1992; Avise, 1994), but this seems less likely in the present ease. One may hypothesize that the single basal clade of the mark phylogeny is the true species phylogeny for section Paeonia and that the two basal clades of the ITS phylogeny result from random sorting of ancestral polymorphisms. It is very unlikely that two such different types of ITS sequences co-existed in the ancestral populations of section Paeonia, which would then have allowed lineage sorting. In other words, the lineage sorting hypothesis may require an extremely long coalescence time, which in turn requires very large effective population size (Pamilo and Nei, 1988; Kreitman, 1991; Hudson, 1992; Moore, 1995). It is, however, impossible to estimate the effective population size here because the divergence rate of ITS sequences is unknown in Paeonia. In any event, it seems that the longer the coalescence time required for the lineage sorting hypothesis, the more likely that hybridization may be involved, especially in some plant groups, such as peonies, where reticulate evolution coupled with polyploidization has been documented. In this context, it seems reasonable to invoke the hybridization hypothesis to explain the other distinct conflicts between the ITS and matK phylogenies Of section Paeonia. Statistical models that may help determine, the significant levels of preference for hybridization vs. lineage sorting, however, are not available. Development of such models is much needed for accurate interpretation of conflicting gene phylogenies (Moore, 1995). Obtaining additional independent gene phylogenies-can also test the hybridization hypothesis, and sequencing studies of low-copy nuclear genes of peonies are in progress.

Hybrid speciation al different ploidy levels

-Allopolyploidy has been considered to be a primary mode for the formation of fertile and stable hybrid species (Grant, 198 1). Frequency of natural hybrid speciation at the diploid level, however, has been controversial (Rieseberg, Carter, and Zona, 1990; Rieseberg, 1991; Wolfe and Elisens, 1994; Rieseberg, Van Fossen, and Desrochers, 1995). In section Paeonia, the species of hybrid origin identified by ITS and cpDNA sequences include six diploid and ten tetraploid species, and three species with both diploid and tetraploid populations (Fig. 7). the proportion of diploids among the hybrids species is unusually high, suggesting that hybrid speciation at the diploid level has been quite successful in peonies. An even more striking phenomenon is the co-existence of diploids and tetraploids in the same species or a group of species with the same origin (Fig. 7). For example, Paeonia broteri (diploid) and P. coriacea (tetraploid), which are endemic to southern Spain and northern Africa, may have the same origin because they share a substitution in the matK phylogeny (Fig. 1). Extensive vegetative reproduction by rhizomes in peonies may have facilitated survival of initial diploid populations of hybrids until they became fertile or polyploidized. Existence of different ploidy levels in the same or very closely related species may eventually lead to reproductive isolation and further speciation. Studies of reproductive biology and the application of more sensitive molecular markers at the populational level are necessary for understanding hybrid speciation in Paeonia.

Molecular evolution in P. cambessedesii (diploid) and P. russi (tetraploid), two endemic species in the western Mediterranean islands, is noteworthy. ITS sequences of P. russi show nucleotide additivity at nine out of ten sites that are variable between P. lactiflora and P. mairei, strongly suggesting that P. russi is derived via hybridization between these two species. Paeonia cambessedesii shows only partial additivity at three of the variable sites, and was considered previously to have the same origin as a group of species with partial additivity of ITS sequences (Fig. 6; Sang, Crawford, and Stuessy, 1995). The matK phylogeny supports the sister group relationship of P. russi and P. cambessedesii, and thus suggests the same origin for both species, i.e., derived through hybridization between P. lactiflora and P. mairei. The result further implies that gene conversion has operated mote rapidly in diploid P. cambessedesii than in tetraploid P. russi. This is reasonable because in diploids, loci of nrDNA are more easily brought together during meiosis, which allows more effective interaction among these loci and thus more rapid gene conversion (Arnheim, 1983). However, this hypothesis does not apply to a group of diploid and tetraploid species, P. clusii, P. rhodia, P. broteri, P. coriacea, P. mlokosewitschii, and two P. mascula subspecies, which may have the same origin and almost identical ITS sequences. In particular, diploid P. broteri and tetraploid P. coriacea, whose common origin is supported by one substitution in matK (Fig. 1), have the same pattern of partial additivity of ITS sequences. The tempo of concerted evolution at different ploidy levels, thus, needs further investigation.


An intercontinental disjunction occurs between section Oneapia, endemic to western North America and the other two sections found in Eurasia. The isolation could have resulted either from a vicariance event disrupting continuous distribution of the ancestral populations between Eurasia and western North America, or a long-distance dispersal from one region to the other. The vicariance explanation is favored here because Paeonia, with follicle fruits and seeds having smooth surfaces and diameters of 7-13 mm, does not appear to have great dispersal ability. Continuous distribution of ancestral Populations of Paeonia between Eurasia and western Northern America is likely to have existed through the Bering land bridge, which allowed periodical exchange of temperate plants between eastern Asia and western North America until late Tertiary or Quaternary (Wolfe, 1975, 1980; Tiffney, 19S5). Distribution of this continuous distribution may have been due to climatic cooling at high latitudes and/or submergence of the Bering land bridge (Tiffney, 1985).

the time of such a vicariance event can be estimated using a molecular clock. Time of divergence may be calculated as the value of DNA sequence divergence divided by twice the sequence divergence rate. For peonies, sequence divergence rates of the sequenced DNA regions are unknown, and cannot be estimated by either fossil records or biogeographic events, We can only use rates estimated in other plant groups. ITS sequences are probably not a good choice for use as a molecular clock because the rates of ITS sequences calculated in several plant groups vary considerably (Suh et al., 1993-, Sang et al., 1995). For matK sequences, the overall divergence rates were suggested as being approximately twice as fast as that of rbcL sequences (Steele and Vilgalys, 1994). lf an average overall divergence rate of 2 x 10-111 per site per year is used for rbcL sequences (Albert et al., 1994), a rate of 4 x 10-10 per site per year can be used for matK sequences. The difference time between section Oneapia and the rest of die genus, thereby, is estimated to be 16.6 million years ago (mya). This estimate, however, is subject to several sources of error. First, the divergence rate of rbcL may not apply to peonies, because it can vary in different groups with different generation times (Li, Tanimura, and Sharp, 1987; Clegg, 1990; Gaut et al., 1992). Vegetative reproduction by rhizomes is very common in peonies, which may significantly prolong generation time, and consequently yield slower rates of DNA divergence. In this case, the divergence time may be underestimated. Second, the estimation that matK evolves twice as fast as rbcL is rough and may actually be different in peonies. Nevertheless, although estimation of divergence rates or times using the molecular clock hypothesis has been based largely on uncertain assumptions and approximate values, it continues to be useful in helping understand tempos of evolution and plant historical biogeography (Parks and Wendel, 1990; Crawford, Lee, and Stuessy, 1992; Wendel and Albert, 1992).

The estimated time for formation of the intercontinental disjunction in peonies, 16.6 mya, is middle Miocene. Tiffney (1985) suggested that during the Miocene, temperatures at higher latitude allowed exchange of deciduous temperate plants via the Bering land bridge. Further, many herbaceous angiosperm groups evolved during the Miocene and exhibited an eastern Asian-eastern North American disjunct distribution (Tiffney, 1985). Formation of intercontinental disjunction in peonies, therefore, may well be a result of disruption of continuous distribution through the Bering land bridge during Miocene time. the possible existence of ancestral populations of extant Paeonia species in high latitudes around the Bering land bridge in the Miocene is concordant with warm climate during this period (Potis and Behrensmeyer, 1992).

Distributional histories of the largest and most widespread section Paeonia may have been much confounded by Pleistocene glaciation (Fig. 8). Reconstruction of complex reticulate evolution within the section provided essential evidence for understanding its biogeography (Sang, Crawford. and Stuessy, 1995). It is striking that after hybridization, European populations of the present Asian species appear to have been completely replaced by their hybrids. Extensive hybridization of peony species must have produced a wide spectrum of different genome combinations upon which natural selection could act (Rieseberg and Wendel, 1993; Arnold and Hodges, 1995). Hybrids that adapted to drastic climatic changes during Pleistocene in Europe are currently distributed in the Mediterranean region. Asian species did not survive such changes in Europe, and their distributions became more restricted. The hybrid species P. xinjiangensis, P. emodi, and P. sterniana may represent footprints of the eastern Asian species P. lactiflora, P veitchii, and P. mairei when their distributional ranges became reduced to only eastern Asia (Fig. 8). Eastern Asia was much less seriously affected by Pleistocene glaciation and may have provided refugia for the nonhybrid peony species (Potts and Behrensmeyer, 1992-, Tao, 1992).

In the Mediterranean region, effects of periodical glaciation on distributions of peony species are speculated. A group of species, P. clusii, P. rhodia, P. broteri, P. coriacea, P. mlokosewitschii, and P. mascula ssp. hellenica and ssp. mascula, endemic to widely different areas in the Caucasus and throughout the southern Mediterranean, may have had a single hybrid origin (Figs. 7, 8). Other than the hypothesis that these hybrid species were dispersed widely across the southern Mediterranean region, a vicariance explanation is again favor-ed. After hybridization, the ancestral hybrid populations might have migrated northward and extended their distributional ranges both eastward and westward during a glacial interval. During subsequent glaciation, these populations were forced into isolated areas in the southern Mediterranean region. Geographic isolation consequently led to speciation among these populations and created the complex diversity now seen for this group of species. A similar distributional pattern is found in another group of Mediterranean species, P. arietina, P. humilis, P. officinalis, and P. parnassica, which also have the same hybrid origin.