Materials and Methods

Plant materials.

The list of wild roses used in this study is provided in Table 1. Sample numbers 1–5, 8–11, 13–24, and 29–46 were grown in the field at the Rose Trial Station of the Shenzhen Park Service, Guangdong, China. Sample numbers 6, 7, 14, 15, 26, and 27 were grown in the field at the Flower Research Institute nursery, Yunnan Academy of Agricultural Sciences, Yunnan, China. Sample numbers 12, 28, 47, and 48 were grown in the field at the Shanghai Chenshan Botanical Garden, Shanghai, China. Morphological traits and literature descriptions were used to confirm the identity of specimen (Ku and Robertson, 2003; Rehder, 1940; Wissemann, 2003). Four species from the related genera Rubus L. and Spiraea L. (Rubus lambertianus Ser., Rubus hirsutus Thunb., Rusbus corchorifolius L., and Spiraea trilobata L.) were chosen as the out-group and samples were collected from Zijin Mountain, Jiangsu, China. Young leaves of all accessions were collected and froze at −80 °C for storage until use. All the voucher samples were preserved and archived in the Herbarium of Nanjing Forestry University.

Table 1.Taxa sampled for rpl16, trnL-F and atpB-rbcL sequence analyses.

View Fullscreen

DNA extraction, polymerase chain reaction (PCR) amplification, and sequencing.

Seven DNA regions from three genomes were examined for their variability among the accessions: the ITS of nrDNA, the rpl16 intron, the trnL-F and atpB-rbcL IGS of cpDNA, and the ccb256, cox2, and nad1 genes of mitochondrial DNA (mtDNA). Among these seven regions, the ITS region was abandoned owing to extensive sequence polymorphisms within a single genotype (Matsumoto et al., 2000; Wissemann and Ritz, 2005; Zhu et al., 2015), and the mtDNA ccb256, cox2, and nad1 genes were abandoned because of a lack of informative sites in the data matrix (0.4%, 0.3%, and 0.6%, respectively, excluding the out-group, unpublished results). The three cpDNA regions were comparatively variable and were used for phylogenetic analyses.

Total DNA from the leaves of individual plants was isolated using a Genomic DNA Extraction Kit (DP305; TianGen Co., Ltd, Beijing, China). The three cpDNA regions were amplified by PCR in a reaction mixture of 25-µL total volume, containing 10 mm Tris-HCl (pH 8.0), 50 mm KCl, 1.5 mm MgCl₂, 200 µm deoxynucleotide triphosphates, 2.5 U Taq-DNA polymerase, 50 ng template DNA, and 1.0 µm forward and reverse primers of rpl16, trnL-F, and atpB-rbcL, accordingly (Table 2). The amplification conditions for each region were as follows: for rpl16, an initial 4 min at 95 °C, followed by 32 cycles of 1 min at 94 °C, 1 min at 52 °C, 1 min at 72 °C, and 10 min at 72 °C for the final extension; for trnL-F, an initial 3 min at 95 °C, followed by 35 cycles of 30 s at 95 °C, 30 s at 50 °C, 1 min 20 s at 72 °C, and 7 min at 72 °C for the final extension (Bruneau et al., 2007); and for atpB-rbcL, an initial 3 min at 95 °C, followed by 30 cycles of 30 s at 95 °C, 45 s at 48 °C, 1 min at 72 °C, and 5 min at 72 °C for the final extension (Wissemann and Ritz, 2005). Amplifications were performed in an Eppendorf thermocycler (Mastercycler Nexus X1). The amplified products were sequenced by Invitrogen (Shanghai, China). Sequencing primers were identical to those used for amplification.

Table 2.Primers for PCR amplification and sequencing.

View Fullscreen

Phylogenetic analyses.

The sequences were aligned using Clustal X (Thompson et al., 1997) and checked manually in BioEdit (Hall, 1999). Our three cpDNA sequence alignments were used to construct separate (not shown in this paper) and combined data matrixes. Modeltest (Posada and Crandall, 1998) was used to perform hierarchical likelihood ratio tests to estimate an optimal nucleotide substitution model for use with MrBayes (Ronquist and Huelsenbeck, 2003). The MP method in PAUP* 4.0b10 (Swofford, 2002) was used with its heuristic search option to find the most parsimonious trees. Insertions and/or deletions (indels) were included in the matrixes, and the resulting gaps were treated as missing data. Character state changes were all weighted equally. Searches were performed using 1000 replicates for bootstrap analysis (Felsenstein, 1985).

BI using Monte Carlo Markov chain (MCMC) was performed with MrBayes 3.0b4 (Ronquist and Huelsenbeck, 2003). Four incrementally heated simultaneous MCMCs were run over 2,000,000 generations, using the general time reversible nucleotide substitution model with gamma distributed substitution rates, random starting trees, and default nucleotide substitution model starting values. Trees were sampled every 100 generations resulting in an overall sampling of 20,001 trees. The first 1000 “burn-in” trees were excluded from each analysis. A 50% majority rule consensus tree was computed to obtain estimates of a posteriori probabilities. Two independent analyses were conducted using these initial parameters.

Results

Sequence analysis.

Summary sequence statistics for the three cpDNA regions were presented in Table 3. Among these three regions, the proportion of informative characters was highest in the rpl16 region (7.6% and 4.6%, depending on inclusion or exclusion of the out-group, respectively). The percentage of informative characters in the trnL-F and atpB-rbcL were 3.5% and 2.4%, within the in-group data matrixes, respectively. These three regions were all relatively easy to align; however, two microsatellites (9–12 A′s and 8–15 A′s) at positions 342–353 and 418–432 in the rpl16 region, and one microsatellite (9–12 T′s) at positions 427–438 in the trnL-F spacer were excluded from analyses data because these regions could not be reliably aligned (Bruneau et al., 2007).

Table 3.Sequence and parsimony analysis statistics for separate and combined chloroplast DNA data sets used in phylogenetic analyses.

View Fullscreen

Phylogenetic analyses.

A total of 2610 nucleotides were used for phylogenetic analyses, heuristic searches of combined data matrix including indels produced 285 variable characters and 207 informative characters, between in-group and out-group in the data matrix (Table 3). The heuristic search resulted in 2000 trees, with a tree length of 607, a consistency index 0.873, and a retention index of 0.902. The Bayesian 50% majority rule consensus of 19,001 trees in stationary phase from the two analyses is presented in Fig. 1. Results from the Bayesian analyses were very similar to those using parsimony.

View Figure

• Download as Powerpoint
• Download Figure

Both methods suggested that the 48 taxa used in our study should be divided into two major groups. The first group was composed of taxa from R. sects. Pimpinellifoliae, Rosa, and Carolinae. Within this group, taxa from R. sects. Rosa and Carolinae formed an inner clade, two other clades were set apart and were constituted by taxa both from R. sect. Pimpinellifoliae. The second group included all the remaining taxa and was further divided into five clades: the first clade consisted of taxa from R. sects. Synstylae, Chinenses, Gallicanae, and Caninae as well as R. ×cooperii (sample no. 43), the molecular data placed taxa from R. sect. Caninae, and R. arvensis (sample no. 28) of R. sect. Synstylae in a sister relationship to the reminder accessions. The second clade consisted of taxa from R. sect. Banksianae, except for R. ×fortuniana (sample no. 37), which formed the third clade together with R. laevigata (sample no. 42) of R. sect. Laegigatae. The fourth clade and the fifth clade contained accessions from R. sects. Microphyllae and Bracteatae, respectively.

Discussion

The 48 taxa phylogeny presented (Fig. 1) based on our three concatenated sequence regions conforms closely to the conventional classification of Rosa botanical sections (Matsumoto et al., 1998; Qiu et al., 2012). For example, five species of the traditional R. sect. Banksianae formed a clade in our phylogeny with high posteriori probability and bootstrap support (100% and 86%, respectively). The phylogenetic results are also consistent with other studies in which taxa from genus Rosa are generally divided into two main clades, named Cinnamomeae (Rosa) clade or Clade I, and Synstylae clade or Clade II (Bruneau et al., 2007; Fougère-Danezan et al., 2015). However, notable exceptions exist.

Three accessions (sample no. 1–3) of R. foetida were separated from the other taxa (sample no. 4–7) of R. sect. Pimpinellifoliae. This result has also been found in previous studies (Bruneau et al., 2007; Fougère-Danezan et al., 2015; Matsumoto et al., 1998). The fertility of tetraploid R. foetida is strikingly low (Roberts, 1977; Rowley, 1967), and it may represent an ancient hybrid of unknown parentage (Wylie, 1954). Our clustering results resolved the other four taxa (sample no. 4–7) of R. sect. Pimpinellifoliae as a sister relationship to the taxa from R. sects. Rosa and Carolinae. While in a more extensive taxon sampling of R. sect. Pimpinellifoliae made by Fougère-Danezan et al. (2015), several species (R. spinosissima L., R. tsinglingensis Pax & K. Hoffm., R. farreri Stapf ex Cox, and R. koreana Kom.) of this section are embedded in the Rosa-Carolinae clade. Moreover, the above four taxa are further divided into two small clades based on our molecular analyses (Fig. 1). In Flora of China (Ku and Robertson, 2003), R. primula (sample no. 4) and R. xanthina (sample no. 5) belong to series Spinosissimae with five petals, whereas R. omeiensis (sample no. 6) and R. sericea (sample no. 7) belong to series Sericeae with four petals. Our molecular results seem to well reflect the morphological peculiarities of these species in R. sect. Pimpinellifoliae. Similar results have also been obtained in previous molecular phylogenetic studies in this genus (Fougère-Danezan et al., 2015; Matsumoto et al., 1998; Qiu et al., 2012).

The Carolinae section should be merged with the Rosa section according to several authors (Bruneau et al., 2007; Fougère-Danezan et al., 2015; Koopman et al., 2008; Matsumoto et al., 1998; Scariot et al., 2006; Takeuchi et al., 2000). This result was confirmed in our analysis. Morphologically R. sects. Rosa and Carolinae can be distinguished by week differences: the sepals are usually upright and persistent in R. sect. Rosa vs. reflected and deciduous in R. sect. Carolinae (Rehder, 1940; Wissemann, 2003). However, the distinction of this character is not substantiated on molecular level in these two sections. R. palustris f. plena (sample no. 48), traditionally placed in the Carolinae section (Table 1), formed a group with all the accessions (samples no. 8–15) of R. sect. Rosa in our study with 100% posteriori probability and 91% bootstrap support (Fig. 1).

Taxa from R. sect. Synstylae were clustered with accessions of R. sect. Chinenses in the phylogeny (Fig. 1). Several authors have already pointed out the close relationship between R. sects. Synstylae and Chinenses (Bruneau et al., 2007; Fougère-Danezan et al., 2015; Jan et al., 1999; Matsumoto et al., 1998, 2000; Millan et al., 1996; Qiu et al., 2012; Wissemann and Ritz, 2005; Wu et al., 2000; Zhu et al., 2015). The morphological similarity of these two sections also concurs with our molecular study, for both sections have adnate and persistent stipules, and exserted styles, taxa of R. sect. Chinenses only lack the connected styles (Ku and Robertson, 2003; Rehder, 1940). It appears these two sections are not monophyletic in the phylogeny in our study (Fig. 1). Similar results have been reported in several previous studies (Bruneau et al., 2007; Fougère-Danezan et al., 2015; Matsumoto et al., 1998; Qiu et al., 2012; Wissemann and Ritz, 2005; Wu et al., 2000; Zhu et al., 2015), but there are also contrary studies suggested that these two sections are independently monophyletic (Jan et al., 1999; Matsumoto et al., 2000; Millan et al., 1996). Moreover, taxa from R. sect. Synstylae can be further divided into series Multiflorae and Brunonianae, respectively, depending on the stipules are pectinate or entire according to Flora of China (Ku and Robertson, 2003). However, this morphological character was not discovered in our study, nor in a recent report by Zhu et al. (2015). Besides, three samples of R. odorata (R. sect. Chinenses, sample no. 31–33) formed an inner clade with R. brunonii (sample no. 20), R. helenae (sample no. 21), and R. weisiensis (sample no. 27) of R. sect. Synstylae, is similar to two recent studies (Fougère-Danezan et al., 2015; Zhu et al., 2015). R. chinensis var. mutabilis (sample no. 29) was first reported in our study, and it presented in the Synstylae-Chinenses clade with an unresolved position (Fig. 1), which may be due to its remarkable variation in the flower color during blooming (Table 1). Nevertheless, more comprehensive molecular studies are still needed to ascertain the position of this peculiar species.

Within the Synstylae-Chinenses clade, R. ×damascena (sample no. 47) from R. sect. Gallicanae was merged with R. sects. Synstylae and Chinenses, while two taxa of R. sect. Caninae, together with R. arvensis (sample no. 28), formed a separate small clade (Fig. 1). The results show that R. sect. Gallicanae, rather than R. sect. Caninae, is a closer relative to R. sects. Synstylae and Chinenses, and that R. sect. Caninae appears to be sister to the Synstylae-Chinenses clade. Several previous studies obtained similar results based on the research on R. gallica L. of R. sect. Gallicanae (Koopman et al., 2008; Millan et al., 1996; Scariot et al., 2006; Wissemann and Ritz, 2005). However, in recent studies, R. gallica formed a clade with taxa from R. sect. Caninae, and both of these two sections appear sister relationship to the Synstylae-Chinenses clade (Fougère-Danezan et al., 2015; Zhu et al., 2015). R. ×damascena is thought to be a hybrid with three parents of wild species, and R. moschata Herrm. of R. sect. Synstylae is considered the female parent (Iwata et al., 2000; Millan et al., 1996). This viewpoint is supported by Takeuchi et al. (2000) whose study showed that R. odorata var. gigantea, R. moschata, and R. ×damascena were grouped into one clade apart from R. sect. Gallicanae.

It is interesting that the only European species of R. sect. Synstylae, R. arvensis, together with R. canina (sample no. 45) of R. sect. Caninae formed a small clade in our analysis (Fig. 1); this result is consistent with previous studies (Wissemann and Ritz, 2005; Zhu et al., 2015), and it also helps to explain why the origin of R. ×alba L. is ambiguous, because they are both hypothetical parents of R. ×alba (Klastersky, 1968; Millan et al., 1996). Furthermore, the origin of R. ×cooperii (sample no. 43) is still in dispute. R. ×cooperii is thought to be a seedling of R. laevigata (Beales, 1985), or a natural hybrid between a seed parent, R. odorata var. gigantea, and a pollen parent, R. laevigata (Young and Schorr, 2007). However, on the basis of our cpDNA molecular data, it formed a clade with R. odorata var. gigantea (Synstylae-Chinenses clade), rather than being clustered with R. laevigata (Fig. 1). This result suggests that the seed parent of R. ×cooperii might not be R. laevigata considering the maternal inheritance of cpDNA in Rosa (Matsumoto et al., 1998). Meanwhile, since this species was also not clustered with R. odotata var. gigantea in the Synstylae-Chinenses clade, it could not be confirmed whether the seed parent of R. ×cooperii is R. odorata var. gigantea, but is possibly a parent from R. sects. Synstylae and Chinenses.

Our clustering results resolved five samples (sample no. 34–38) of R. sect. Banksianae as monophyletic and as sisters to the Synstylae-Chinense clade, without considering the position of R. ×fortuniana (sample no. 39). Similar results were obtained in previous studies (Fougère-Danezan et al., 2015; Qiu et al., 2012; Wu et al., 2001). R. ×fortuniana, which is traditionally placed in the R. sect. Banksianae, formed a clade with R. laevigata with 100% posteriori probability and 100% bootstrap confidence in our plastid sequence analyses (Fig. 1). R. ×fortuniana is thought to be a hybrid between a R. laevigata seed parent and a R. banksiae pollen parent, and its cpDNA matK sequence matches that of R. laevigata, not R. banksiae (Matsumoto et al., 2001). This matK sequence result is consistent with our molecular data. R. sects. Laevigatae and Bracteatae were also resolved as sister relationship to the Synstylae-Chinense clade in the second group in our study. Similar results were indicated in other studies for R. sect. Laevigatae (Jan et al., 1999; Zhu et al., 2015), for R. sect. Bracteatae (Wissemann and Ritz, 2005), or for the both sections (Fougère-Danezan et al., 2015; Qiu et al., 2012). Furthermore, R. sects. Banksianae, Laevigatae, and Bracteatae are morphologically united in a large clade according to the free and deciduous stipules (Rehder, 1940). However, R. sect. Banksianae discriminates from R. sect. Bracteatae (which is readily featured by its woolly or tomentose young receptacle and branchlet) by the nonpubescent receptacle and branchlet, and discriminates from R. sect. Laevigatae by the reflexed and deciduous sepals. In addition, the stipules of R. sect. Bracteatae are deeply incised, which obviously differs from the other two sections. Several previous analyses also indicate that these three sections are distantly related to each other (Fougère-Danezan et al., 2015; Jan et al., 1999; Qiu et al., 2012). So it seems that the morphological characterization of free and deciduous stipules in these three sections indicates not synapomorphy but convergence or plesiomorphy (Wissemann and Ritz, 2005). Besides, R. roxburghii (sample no. 40) and R. roxburghii f. normalis (sample no. 41) formed a clade, and in parallel with R. sects. Banksianae, Laevigatae, and Bracteatae (Fig. 1). This agrees with previous suggestions that R. roxburghii should be placed on a sectional level (Crépin, 1889; Fougère-Danezan et al., 2015; Jan et al., 1999; Koopman et al., 2008; Ku and Robertson, 2003; Wissemann and Ritz, 2005), rather than being classified as a subgenus (Rehder, 1940; Wissemann, 2003).

In summary, our phylogenetic reconstruction (Fig. 1) derived from the combined three cpDNA data for 48 genotypes shows substantial consistency with the conventional classification of infrageneric sections (Table 1). However, the results are mostly not consistent with the conventional taxonomy in a better sampling of both taxa and molecular markers (Fougère-Danezan et al., 2015; Zhu et al., 2015). The differences might be caused by the hybridization (genetic introgression) and/or polyploidization existing among the taxa of this genus. For instance, R. lichiangensis (sample no. 26) has been confirmed to be a hybrid species (Zhu et al., 2015), and the decaploid R. praelucens Byhouwer is also regarded as an allopolyploid resulting from hybridization (Fougère-Danezan et al., 2015; Jian et al., 2010). Moreover, chloroplast markers ignore hybridization (because of their maternal inheritance), and concerted evolution and homogenization of parental polymorphism exist in the ITS region of nrDNA. Therefore, low/single copy nuclear genes, such as those used by Zhu et al. (2015), should be employed in future studies to better understand this complex genus.