| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
What's this? |
Systematics and Phytogeography |
2 Plant Research International, Wageningen UR, P.O. Box 16, 6700 AA Wageningen, The Netherlands 3 Institut fuer Allgemeine Botanik und Pflanzenphysiologie AG Spezielle Botanik, Justus-Liebig-Universitaet Giessen, Senckenbergstr. 17 35390 Giessen, Germany 4 Unit Plant, Applied Genetics and Breeding, Institute for Agricultural and Fisheries Research, Caritasstraat 21, 9090 Melle, Belgium 5 Department of Systematic Botany, Friedrich Schiller Universität Jena, Philosophenweg 16 07743 Jena, Germany 6 Ecologisch Adviesbureau Maes, Achter Clarenburg 2, 3511 JJ Utrecht, The Netherlands 7 Balsgård-Department of Plant Breeding and Biotechnology, Swedish University of Agricultural Sciences, Fjälkestadsvägen 459, 291 94 Kristianstad, Sweden 8 Institute for Plant Genetics, Leibniz Universität Hannover, Herrenhäuserstrasse 2 30419 Hannover, Germany
Received for publication 29 June 2007. Accepted for publication 31 January 2008.
ABSTRACT
The genus Rosa has a complex evolutionary history caused by several factors, often in conjunction: extensive hybridization, recent radiation, incomplete lineage sorting, and multiple events of polyploidy. We examined the applicability of AFLP markers for reconstructing (species) relationships in Rosa, using UPGMA clustering, Wagner parsimony, and Bayesian inference. All trees were well resolved, but many of the deeper branches were weakly supported. The cluster analysis showed that the rose cultivars can be separated into a European and an Oriental cluster, each being related to different wild species. The phylogenetic analyses showed that (1) two of the four subgenera (Hulthemia and Platyrhodon) do not deserve subgeneric status; (2) section Carolinae should be merged with sect. Cinnamomeae; (3) subsection Rubigineae is a monophyletic group within sect. Caninae, making sect. Caninae paraphyletic; and (4) there is little support for the distinction of the five other subsections within sect. Caninae. Comparison of the trees with morphological classifications and with previous molecular studies showed that all methods yielded reliable trees. Bayesian inference proved to be a useful alternative to parsimony analysis of AFLP data. Because of their genome-wide sampling, AFLPs are the markers of choice to reconstruct (species) relationships in evolutionary complex groups.
Key Words: amplified fragment length polymorphism (AFLP) • Bayesian inference • parsimony • phylogeny • Rosa • Rosaceae • roses • UPGMA
The history of rose systematics can best be characterized by the saying "where there are two systematists, there are three opinions." Starting with the first modern classification by Linnaeus in 1735, the number of Rosa species recognized ranged from 12 to many hundreds (see Wissemann [2003]
for an overview). Currently, there is a broad consensus for using the system of Rehder (1940), which was recently updated by Wissemann (2003). The updated system comprises four subgenera: Hulthemia (1 species), Rosa (some 180 species), Hesperhodos (2 species), and Platyrhodon (1 species). Subgenus Rosa is subdivided into 10 sections, and the largest of these sections (sect. Caninae) is again subdivided into six subsections.
Most of the taxonomic confusion in the genus Rosa originates from the complicated evolutionary history of the wild species, combined with a long history of cultivation and interbreeding of selected genotypes. The complexity is caused by several factors, often in conjunction: (1) extensive hybridization, both ancient and recent (Wissemann and Ritz, 2005; Joly et al., 2006); (2) absence of clear differences between many of the species, partly due to their recent radiation (Wissemann and Ritz, 2005); (3) incomplete lineage sorting (a common feature in recently diverged species; Joly et al., 2006); and (4) polyploidy (with multiple/hybrid origins for the polyploids in at least some of the species; Joly et al., 2006).
A second source of confusion is the use of morphology as the basis for the Rosa classifications. One of the requirements for a reliable phylogenetic character is that the mode and direction of character evolution is representative for the mode and direction of the species evolution as a whole. Morphological characters are often under severe selection pressure, for example when growth conditions (rapidly) change. The selection pressure may on the one hand result in character similarity for evolutionary divergent species adapting to similar conditions, and on the other hand in striking morphological differences between related species adapting to different conditions. The genus Rosa contains examples of both.
An example of character similarity is connected to subsect. Rubigineae. This subsection can be characterized by a granular type of epicuticular wax, whereas members of the other subsections have triangular rodlets. However, three less-closely related species outside subsect. Rubigineae also have the granular wax type (Wissemann, 2000). Divergent character evolution is present in R. persica, having a unique leaf morphology with a single leaflet without stipules. The unique morphology gave rise to much taxonomic controversy, with R. persica sometimes being excluded from the genus Rosa. However, recent insights clearly have shown that R. persica is a member of the genus Rosa, notwithstanding its deviating leaf morphology (Wu et al., 2001b; Wissemann and Ritz, 2005). The examples show that morphology can be a misleading indicator of species relationships in Rosa and that an alternative source of phylogenetic information is needed.
In the 1990s, molecular markers were developed for rose cultivar identification (see Debener et al. [1996] and Millan et al. [1996] for overviews), and several of these were tested for identifying species relationships in Rosa. Debener et al. (1996) and Millan et al. (1996) used RAPDs to examine the relationships among cultivars and a limited number of wild species. The work of Debener et al. (1996) showed a distinction between a group of cultivars and a group of wild species. In a study on wild accessions, Millan et al. (1996) showed a clustering largely according to the sectional affinities. The latter was also found by Jan et al. (1999). Wu et al. (2001a) used RAPDs to study the relationships within sect. Synstylae, but obtained a tree with little resolution. More recently, Wen et al. (2004) employed RAPDs to study the relationships of R. roxburghii and relatives, and Bruneau et al. (2005) used them to study R. blanda and segregates. However, problems with the reproducibility of RAPD markers (discussed in e.g., Jones et al. [1997] and Bagley et al. [2001]) may render them less suitable as phylogenetic markers, although there has been some improvement in recent years (Nybom, 2004). Mitochondrial and chloroplast RFLPs were used by Matsumoto et al. (1997) and Takeuchi et al. (2000) to study the relationships among wild Rosa species. Although they were able to draw conclusions on certain groups, the plastid RFLP data showed a general lack of resolution.
As an alternative to restriction fragment markers, several authors examined the use of sequences as a source of phylogenetic information in the genus Rosa. MatK sequences were used by Matsumoto et al. (1998), Wu et al. (2000), and Leus et al. (2004). Although the resolution obtained with these sequences was higher than that with RFLPs (Matsumoto et al., 1998), the resulting phylogenetic trees still suffered from a lack of resolution, and most branches had low bootstrap (Felsenstein, 1985) support. The same was true for trees based on the atpB-rbcL intergenic spacer (IGS; Wissemann and Ritz, 2005). Leus et al. (2004) examined rbcL sequences, but this locus was entirely discarded from further analysis due to a lack of polymorphisms. Bruneau et al. (2007) tested eight noncoding chloroplast regions for their potential as phylogenetic markers. Among these, four regions did not contain sufficient variation for an analysis, while the variation in the remaining four was only limited. Several authors used ITS sequences as a basis for Rosa phylogenies, but the presence of different ITS copies in single individuals due to nonconcerted evolution (Wissemann, 1999, 2002; Ritz et al., 2005Wissemann and Ritz, 2005;) hampered an unambiguous interpretation of the results. Matsumoto et al. (2000) and Wu et al. (2001b) handled the nonidentical paralog sequences by merging the various copies into one representative sequence per individual. Leus et al. (2004) used direct sequencing and included the polymorphisms in the analyses. However, these strategies may introduce considerable error in the phylogenies, as was demonstrated by the distinct placement of different ITS copies from single individuals in the phylogenetic trees of Wissemann (2002) and Wissemann and Ritz (2005). On top of that, the phylogenies inferred from ITS data showed a lack of resolution and low bootstrap support (Matsumoto et al., 2000; Wu et al., 2001b; Wissemann, 2002; Leus et al., 2004; Wissemann and Ritz, 2005). Recent work in North American roses shows promising results for glyceraldehyde 3-phosphate dehydrogenase (GAPDH), triose phosphate isomerase (TPI), and malate synthase (MS) gene sequences when allelic variation is taken into account (Joly and Bruneau, 2006; Joly et al., 2006), but this approach has not yet been tested extensively.
To overcome the drawbacks of RAPDs and RFLPs, alternative markers were developed, most notably microsatellites or simple sequence repeats (SSRs) (reviewed in Cregan, 1992), and AFLPs (Vos et al., 1995). Both marker types combine high reproducibility with high variability, potentially increasing both reliability and resolution of phylogenies. On top of that, both AFLPs and microsatellites enable a genome-wide sampling, increasing the chance that the data sets and phylogenies represent the evolutionary affinities of the species rather than that of the individual characters.
Concerning microsatellites, a marker set originally developed for cultivar identification (Esselink et al., 2003) proved to be useful for identifying relationships among cultivars and species, yielding trees with considerable support and resolution (Scariot et al., 2006). However, the phylogenetic interpretation of microsatellite data is complicated because of allele-size constraints, the uncertain and poorly understood mechanisms of allele-size change, and the uncertain influence of mutational events on allele length (Page and Holmes, 1998).
Phylogenetic interpretation of AFLP markers may be more straightforward, as was first demonstrated in a study in Solanum by Kardolus et al. (1998). In Rosa, AFLP data were first tested in a number of smaller studies demonstrating their use for detecting relationships among cultivars and species in Rosa. For example, in a set of 88 rose genotypes, Leus et al. (2004) detected a clear distinction between cultivars and wild species. Comparable results were obtained by Wen et al. (2004) in a study on seven cultivars and seven wild relatives of R. roxburghii. Bruneau et al. (2005) used AFLPs to examine the species boundaries among R. blanda and four closely related species and were able to demonstrate that two of them were conspecific with R. blanda. Deák et al. (2005) studied the genetic relationships among 34 taxa of Rosa from Hungary (including many hybrids), and found that the main groups in their AFLP dendrogram largely agreed with the commonly recognized sections. Joly and Bruneau (2007) successfully used AFLPs to delimit the species boundaries in Rosa sect. Cinnamomeae.
Given the drawbacks of RAPDs, RFLPs, sequences, and microsatellites, and the promising results obtained with AFLPs, the latter seem to be a suitable type of markers for inferring relationships in Rosa. However, AFLPs have their own specific drawbacks, most notably (1) possible nonindependence of fragments, (2) problems of homology assignment of fragments, (3) asymmetry in the probability of losing and gaining fragments, and (4) problems in distinguishing heterozygote from homozygote bands (reviewed in Koopman, 2005). Theoretically, these drawbacks can obscure the phylogenetic signal to such an extent that a reliable phylogeny reconstruction becomes impossible. For a data set of Lactuca species, Koopman (2005) showed that in practice, sufficient signal remains to enable a reliable phylogeny reconstruction. However, the Lactuca data set contained only diploid species, and there was no indication of extensive hybridization among them.
In Rosa, with its extensive hybridization, absence of clear species differences, and polyploidy, the situation is far more complex than it is in Lactuca. Although the smaller studies described earlier indicate at least some use for AFLPs in Rosa phylogeny, they all concern phenetic analyses of a limited set of species. The current study is the first to analyze a wider sample of Rosa species both phenetically and phylogenetically, and we are among the first to examine the use of Bayesian inference for the phylogenetic analysis of AFLP data. (Brouat et al., 2004; Vriesendorp, 2007). The value of AFLPs for determining species relationships in Rosa is evaluated by comparing the present results with those of previous morphological and molecular studies. The value of Bayesian analysis for phylogeny reconstruction from AFLP data sets is evaluated by comparison with a parsimony analysis of the same data.
A complicated evolutionary history is not limited to the genus Rosa, but is common to many species groups. Therefore, the present results may also contribute to a better understanding of AFLP markers as a source of phylogenetic information in other complex groups and of the merits of Bayesian vs. parsimony analysis of AFLP data in such groups.
MATERIALS AND METHODS
Plant materials
We used 92 accessions from 46 species of the genus Rosa, comprising three of four subgenera, six of 10 sections of sect. Rosa, and all six subsections of sect. Caninae. To minimize the risk of including spontaneous hybrids, we limited the sampling from the Wageningen botanical gardens to materials that originated directly from the wild or that descended directly (i.e., by only one round of propagation) from wild-collected material at other botanical gardens. The species accessions were supplemented with 33 cultivars from various cultivar groups, comprising a wide sample of present day cultivar variation. The accessions are listed in Appendix S1 (see Supplemental Data with online version of this article). All accessions from Appendix S1 were included in a phenetic analysis of species relationships and cultivar affinities.
Rose cultivars typically result from complex crosses among older cultivars and/or wild species, and therefore they must be regarded as hybrid taxa. Most phylogeny reconstruction methods are not fit for hybrid taxa (see Data analysis section), and therefore the cultivars (marked with a double asterisk in online Appendix S1) were excluded from the phylogenetic analyses. We made an exception for one Rosa xcentifolia and one Rosa xdamascena cultivar. Rosa xcentifolia and Rosa xdamascena cannot be represented by wild accessions, because they consist entirely of cultivated hybrid material and are unknown from the wild. However, they are very important in the ancestry of many present day cultivars and are often treated as species in literature (e.g., Wylie, 1954). Therefore, we decided to retain them in the phylogenetic analyses, notwithstanding their hybrid nature.
DNA extraction
Young leaves were collected and stored on silica gel. DNA was extracted using a Qiagen DNeasy Plant Mini Kit (Westburg, Leusden, The Netherlands) according to the manufacturer's instructions. The extracted DNA was stored at –80°C until use.
AFLP analysis
We used seven primer combinations, selected from a large number of primer combinations used by Yan et al. (2005) to generate polymorphic markers for a genetic map in diploid roses. The primer combinations were selected based on the total number of bands, the clarity of the pattern, and the distribution across the gel.
The AFLP method was performed essentially as described by Vos et al. (1995) with minor modifications (Arens et al., 1998; Smulders et al., 2000). Three hundred nanograms of DNA was digested with restriction enzymes EcoRI and Mse I (Invitrogen Life Technologies, Breda, The Netherlands). Digestion and ligation was performed in a single reaction for 4 h at 37°C. The preamplification was performed with EcoRI/MseI primer pairs, each primer containing one selective nucleotide (EcoRI+A, MseI+C). Final amplifications were performed with three selective nucleotides on each primer.
For the selective amplification of primer combinations E33/M53 (selective nucleotides AAG/CCG) and E33/M54 (AAG/CCT), the EcoRI primer was labeled with 33P. Reaction products were loaded on a 6% polyacrylamide gel (Sequagel-6, Biozym TC, Landgraaf, The Netherlands) in 1x TBE (Tris-borate EDTA) electrophoresis buffer using a SequiGen 38 x 50 cm gel apparatus (Bio-Rad, Nazareth Eke, Belgium). Gels were dried on Whatmann 3MM paper, and x-ray films (Kodak X-OMAT, Rochester, New York, USA) were exposed for 1–3 wk at room temperature. The films were scanned and scored using QuantarPro (Keygene, Wageningen, The Netherlands).
For primer combinations E33/M49 (AAG/CAG), E44/M51 (ATC/CCA), E33/M58 (AAG/CGT), E44/M48 (AAG/CCG), and E33/M50 (AAG/CAT), the Eco RI adapter primers were labeled with IRD700 and IRD800. The fragments were separated on a LI-COR Global Edition IR2 system (LI-COR, Lincoln, Nebraska, USA). Fragments of E33/M50 were detected using SAGAMX AFLP software (LI-COR). Fragments for the remaining primer combinations were detected using QuantarPro (Keygene). Fragments were scored in a length range from approx. 60–515 base pairs. The scores for all primer combinations were compiled into a single binary data matrix (presence = 1, absence = 0) for final analysis.
Data analysis
Numerous tree building methods are available today for the analysis of species relationships, each method having its own advantages and limitations. The most commonly used methods can be divided into clustering methods (e.g., UPGMA, WPGMA, single-link, complete-link), parsimony methods (e.g., Wagner parsimony, Fitch parsimony, Dollo parsimony, neighbor joining), and likelihood methods (e.g., maximum likelihood and Bayesian inference). We analyzed our data using one method from each of these three approaches.
Relative to parsimony and likelihood methods, clustering methods (see Rohlf [1993] for an overview) have the advantage that the trees are calculated without the assumptions of an evolutionary model. Because hybrids usually violate such assumptions, clustering methods are very suitable for calculating trees that include hybrids. Among the clustering methods, UPGMA generally yields trees that best represent the data (i.e., the highest cophenetic correlation; Mace et al., 1999a, Mace et al., 1999b; Koopman et al., 2001). Therefore, we examined the relationships among the cultivars and the wild species using UPGMA clustering. The analysis was performed with the program Treecon version 1.3b (Van de Peer and De Wachter, 1994), based on Nei and Li (1979) distances. Branch support values were determined in 2000 bootstrap replicates.
A drawback of UPGMA (and other clustering methods) is that the resulting trees depict phenetic relationships (reflecting similarity) rather than phylogenetic relationships (reflecting evolution). However, most of the available phylogeny reconstruction methods use evolutionary models that do not apply to hybrids. Therefore, we performed the phylogenetic analyses without the cultivars (which are hybrids), although Rosa xcentifolia and Rosa xdamascena were retained in the analyses (see Plant materials).
As an alternative to tree building methods, several so-called network reconstruction methods are available. In contrast to tree building methods, network reconstruction methods allow for nondichotomous splits between the taxa, thus enabling the reconstruction and representation of reticulate evolutionary patterns. As such, these methods could be an alternative for tree building methods in groups with hybrid taxa. However, the presently available network reconstruction methods are typically designed for use at the population level, and their added value relative to tree reconstruction methods is still uncertain (Vriesendorp and Bakker, 2005). Therefore, the current study is limited to tree building methods.
The basic assumption of the parsimony approaches is that the correct evolutionary tree is the one that requires the smallest number of evolutionary steps to explain the data (the so-called most parsimonious tree). A disadvantage of neighbor-joining (Saitou and Nei, 1987) is that the trees are reconstructed from a distance matrix rather than directly from the original data matrix, which may result in a loss of information. Furthermore, although the neighbor-joining algorithm optimizes the individual branches in the tree, there is no guarantee that the final tree will be the overall shortest one (Hillis et al., 1996). In contrast, the parsimony algorithms optimize the trees directly on the data set and are able to retrieve the overall shortest tree. The various parsimony methods (see Swofford and Begle [1993] for an overview) differ in the character state changes that are allowed. Fitch parsimony allows free changes among multistate character states, Wagner parsimony requires binary or ordered multistate characters, and Dollo parsimony permits derived character states to originate only once (Swofford and Begle, 1993). Given the unequal chances of gain and loss of bands, Dollo parsimony seems to apply to AFLP data. However, a previous study on Lactuca AFLP data (Koopman, 2005) showed that Dollo parsimony is too restrictive to retrieve the correct tree topology. Given the restrictions in Dollo parsimony and the fact that we have binary characters, we analyzed our data using Wagner parsimony.
The parsimony analysis on the entire data set was conducted in the program PAUP* version 4.0b10 (Altivec) (Swofford, 1999) as a heuristic search comprising 100000 random addition sequences (holding one tree at each step), tree-bisection-reconnection (TBR) branch swapping, and MulTrees switched off. Parsimony settings were acctran and collapse of zero length branches. Additional branch swapping on the most parsimonious trees was performed with MulTrees in effect, and the remaining settings as before. Eriksson et al. (2003) showed the genus Rubus to be most closely related to Rosa. Therefore, three Rubus species (R. phoenicolasius, R. fruticosus s.l., and R. caesius) were used as the outgroup. Branch support values were determined in a bootstrap analysis comprising 10000 replicates, each replicate consisting of 10 random addition sequences, and the remaining settings as before. Phylogenetic signal in the data set was quantified with the g1 statistic (Hillis and Huelsenbeck, 1992) from the tree-length distribution skewness (TLD) test as implemented in PAUP* 4.0b10 (Altivec).
Among the likelihood-based methods, maximum likelihood (Felsenstein, 1981) and Metropolis-coupled Markov chain Monte Carlo Bayesian inference (Yang and Rannala, 1997) are the most widely used. Whereas maximum likelihood retrieves the tree that maximizes the probability of observing the data given that tree, Bayesian inference retrieves the tree(s) that maximizes the probability of the tree, given the data. Relative to maximum likelihood, the Metropolis-coupled Markov chain Monte Carlo Bayesian inference has the advantage that it will not accidentally retrieve a suboptimal tree, and that the calculations are expected to be much faster (Leaché and Reeder, 2002). Relative to parsimony, Bayesian inference has the advantage that it allows nonparsimonious solutions (although viewing from the parsimony perspective this may be considered a drawback) and that the evolutionary model can be specified, which may make the analysis more realistic (Lewis et al., 2005). A disadvantage is that the restriction site model available in the phylogenetic software (MrBayes 3.1.2; Huelsenbeck and Ronquist, 2001; Ronquist and Huelsenbeck, 2003; Ronquist et al., 2005) is designed for restriction site data in general and not specifically for AFLPs. It is a simple F81-like (Felsenstein, 1981) model with an instantaneous rate matrix that only takes into account the asymmetry in band gains and band losses (Ronquist et al., 2005). As such, it is a gross oversimplification of the wide range of evolutionary processes underlying AFLP polymorphisms (Luo et al., 2007). The Bayesian inference approach, using a more sophisticated model of AFLP evolution as proposed by Luo et al. (2007), turned out to be about 40000 times slower than the Bayesian inference using the restriction site model in MrBayes. Given this computational burden, the model of Luo et al. (2007) is at present not practically applicable. Besides the limitations to the evolutionary modeling, another disadvantage of Bayesian inference is that it may erroneously attach high posterior probabilities to short branches or to branches supported by a small number of characters (Alfaro et al., 2003; Lewis et al., 2005). Given the possible advantages of Bayesian inference and the fact that as yet little information is available on its use for the phylogenetic analysis of AFLP data (Brouat et al., 2004; Vriesendorp, 2007), we decided to test Bayesian inference as an alternative for Wagner parsimony.
The restriction site model in MrBayes offers a number of options to correct for coding bias (Ronquist et al., 2005). We used the option noabsencesites, which most closely fits the bias in AFLP data sets. AFLPs are typically applied in relationship studies as anonymous markers, generated without any detailed knowledge of the genome. Therefore, loci that do not yield any AFLP fragments go undetected. As a result, it is not possible to distinguish empty band positions resulting from a total absence of the "1" allele (while the locus as such is present), from empty band positions due to the absence of the locus itself. This lack of distinction makes empty band positions an unreliable source of information, and they are usually removed from the data sets. As a result, the coding in AFLP data sets (including our Rosa data set) is typically biased in that it includes only those loci for which a fragment is present in at least one of the genotypes. The coding bias option "noabsencesites" accounts for this bias.
The Bayesian analysis consisted of two runs of 120 000 000 generations each, each run consisting of 10 independent chains, with a temperature of 0.005 for the heated chain, a sample frequency of 5000, and a burnin of 19 000 samples. The burnin value determines how many trees from each run will be discarded prior to the calculation of the posterior probability values. The choice of the burnin value was based on the criterion that stable log likelihood values had to be reached for the sampled trees, and stable convergence values (i.e., average standard deviations of split frequencies) for the two independent runs. With runs of 120 000 000 generations and a sample frequency of 5000, a total of 24 000 trees were sampled from each run, and 19 000 of these trees were discarded as burnin. Hence, the posterior probabilities were calculated based on the remaining 5000 trees per run. Compared to the average bootstrap sample in phylogenetic studies being 1000 or even 100 replicates, we consider this a sufficiently large sample. The Dirichlet prior for the state frequencies was set to (3.00, 1.00), matching the actual 0/1 frequencies in the data set. Rubus phoenicolasius, R. fruticosus s.l., and R. caesius were used as the outgroup.
RESULTS
Phenetic relationships and cultivar affinities
The scoring of all seven primer combinations resulted in a data set with a total of 520 polymorphic bands (on average 123 per genotype), all of which were included in the analysis. The phenetic relationships among the wild species and the cultivars resulting from this analysis are depicted in Fig. 1. The data matrix and the phenogram are available in TreeBASE (http://www.treebase.org).
|
The bulk of the cultivars with a mixed species background (the additional cultivars in Appendix S1, see online Supplemental Data) can also be subdivided into two major clusters, although both clusters are poorly supported: a Rosa cluster and a Synstylae cluster. The Rosa cluster (2a) contains 18 cultivars, including ‘Boule de Nanteuil,' which can be assigned to wild species R. gallica, and Rosa xdamascena ‘Ispahan.' The cultivars in this cluster are most closely related to cluster 2b, containing a mixture of species from three different sections: Rosa (R. gallica), Caninae (R. tomentella and R. turcica), and Synstylae (R. arvensis). The next closest affinity of the Rosa cluster is to the Caninae cluster. The Synstylae cluster (3a) also contains 18 cultivars, in addition to the sect. Synstylae species R. moschata. The cluster is related to a cluster with the wild species R. wichurana and R. multiflora and their cultivars (3b).
Phylogenetic relationships as detected with parsimony
The phylogenetic data set contained a total of 483 variable bands, 421 of which were parsimony-informative. The g1 statistic from the TLD test was –0.22. The data set contains 83 accessions and 483 variable characters, corresponding to a critical value of –0.08 (Hillis and Huelsenbeck, 1992). The g1 value is considerably lower than this critical value, indicating the presence of significant phylogenetic signal in the data set. The parsimony analysis resulted in four most parsimonious trees (MPTs) of 3276 steps. Additional branch swapping on these trees resulted in four more MPTs. The MPTs had a CI (Kluge and Farris, 1969) of 0.147 and an RI (Farris, 1989) of 0.495, both including parsimony-uninformative characters. One of the trees is depicted in Fig. 2. The data matrix and the MPT of Fig. 2 are available in TreeBASE (http://www.treebase.org).
|
Phylogenetic relationships as inferred with Bayesian inference
The log likelihood values of the runs (Appendix S2, see Supplemental data with online version of this article) decreased sharply until generation 15 000 (–LnL run 1 = 12 464; –LnL run 2 = 12 561), and much more slowly between generations 15 000 and 22 000 000 (–LnL run 1 = 12 196; –LnL run 2 = 12 180). Between generations 22 000 000 and 95 000 000, the decrease in log likelihood was only marginal, but occasionally trees with a relatively low log likelihood (–LnL 
12 300) were sampled from the second run. The outliers disappeared after generation 95 000 000, and the log likelihood values kept oscillating around a value of –LnL 12 173 (online Appendix S2). The convergence value (Appendix S3 in Supplemental Data with online version of this article) decreased sharply in the first 1 00 000 generations (to 0.090), more slowly between generations 1 00 000 and 50 000 000 (to about 0.029), and even more slowly after that. The search was aborted in generation 120 000 000 on a convergence value of 0.025. Additional runs with different temperatures and tuning parameters did not result in a better convergence.
The tree with the highest likelihood (–LnL = 12 126) was sampled in generation 24 755 000 of run 2 (Fig. 3). The topology of the tree is largely congruent with that of the MPT (Fig. 2). The data matrix and the tree are available in TreeBASE (http://www.treebase.org).
|
The posterior probabilities in Fig. 3 are generally higher than the bootstrap values for the corresponding clades in Fig. 2, but the posterior probabilities and bootstrap values are not tightly correlated. For example, a posterior probability of 1.00 corresponds to a bootstrap value of 97% in the clade with R. arvensis 2/18/204, but also to a value of 52% in the species pair R. corymbifera 17/R. canina 69 (compare Figs. 2 and 3). A bootstrap value of 71% corresponds to a posterior probability of 0.75 in R. glauca 163/211, but also to one of 0.99 in R. arkansana 445/R. blanda 447.
DISCUSSION
Cultivar affinities
In the phenetic tree of Fig. 1, the cultivars group in two distinct clusters, separated from most of the wild species. Such a separation between cultivars on the one hand and wild species on the other hand agrees with previous studies using RAPDs (Debener et al., 1996; Wen et al., 2004), AFLPs (Leus et al., 2004; Wen et al., 2004), and microsatellites (Scariot et al., 2006). Apparently, the cultivated gene pool is relatively isolated from the wild gene pool. This relatively isolated position is in line with the common practice of breeding new varieties by interbreeding of existing varieties, with only a limited use of wild germplasm. Our results show that the segregation of cultivated and wild accessions is not complete because several wild species cluster with the varieties. This incomplete segregation is also apparent in the results of Scariot et al. (2006) and may be a direct indication of a limited use of wild germplasm in the cultivated gene pool.
According to Wylie (1954), the wild ancestors of the garden roses are the sect. Synstylae species R. moschata, R. wichurana, and R. multiflora; the sect. Gallicanae (=Rosa) species R.xdamascena and R. gallica; the sect. Chinenses (=Indicae) species R. chinensis and R. gigantea; and the sect. Pimpinellifoliae species R. foetida. In our cluster analysis (Fig. 1), cluster 2a is associated mainly with sect. Rosa species and cluster 3a with sect. Synstylae species. The wild species R. foetida clusters in the lower part of the tree at a considerable distance from the cultivar clusters. These results indicate that the contribution of R. foetida to the cultivated gene pool was less extensive than that of the species from sections Rosa and Synstylae. The limited contribution of R. foetida may be explained by the low fertility of R. foetida and its related hybrids, making it a difficult parent in breeding programs. Nevertheless, it was an invaluable source of the yellow flower color (Wylie, 1954). No information was obtained on the sect. Chinenses-Indicae species R. chinensis and R. gigantea, which were not included in the current study.
In the sect. Rosa cluster (2a), R.xdamascena ‘Ispahan' and R. gallica ‘Boule de Nanteuil' cluster with the other cultivars, while a cluster of species including a wild R. gallica provenance (2b) is sister to this group. Therefore, we propose that the Rosa cluster has a mainly European genetic background. In line with this view, the cultivars in cluster 2a belong to the Damask, Centifolia, Gallica, Alba, Moss, and Portland cultivar groups (online Appendix S1). According to Wylie (1954), these cultivar groups all have their main ancestry in the old European garden roses of the sect. Gallicanae (=Rosa), notwithstanding the fact that many of the groups also include hybrid influences of the China roses. The principal involvement of European roses in cluster 2a is further corroborated by the (more distant) affinity with sect. Caninae (cluster 1 in Fig. 1). Rosa canina was suggested as the female parent of the Alba roses (Wylie, 1954). Given their positions in clusters 2a/b, we propose that R. xdamascena was the most important direct ancestor of the cultivars in the Northwest-European (Rosa) cluster, while R. gallica was a slightly more distant ancestor.
In the sect. Synstylae cluster (3a), R. moschata clusters among the cultivars, while a cluster with R. wichurana and R. multiflora is sister to this group. Given these affinities, we propose that this cluster has a mainly Oriental genetic background. The position of R. moschata among the cultivars and R. wichurana/R. multiflora as a sister group indicates that R. moschata contributed most to the genetic composition of the cultivars, while R. wichurana and R. multiflora contributed less. The Oriental background of the cultivars in the Synstylae cluster is in line with the fact that they mainly belong to the Moschata, Multiflora, Noisette, Bourbon, Tea, and Polyantha cultivar groups (online Appendix S1). According to Wylie (1954), these groups show a major influence of 19th and 20th century China rose introductions. The Hybrid perpetuals cultivar group, in our study represented by ‘La Reine' and ‘Yolande d'Aragon', were derived from crosses between China rose hybrids and R. gallica/R. xdamascena hybrids (Wylie, 1954). These cultivars group in cluster 2a of Fig. 1, indicating a predominant R. gallica/R.xdamascena background.
Phylogenetic relationships among the subgenera of genus Rosa
Three of the four subgenera of the genus Rosa are represented in our data set: Hulthemia (R. persica), Platyrhodon (R. roxburghii), and Rosa (remaining species). In the classifications of Rehder (1940) and Wissemann (2003), these subgenera (as well as the fourth subgenus Hesperhodos) are separated by distinctive autapomorphic morphological characters. In the MPT (Fig. 2), R. persica is sister to R. foetida, and in the tree from the Bayesian analysis R. persica is the earliest derived species. R. roxburghii is sister to R. hugonis in the MPT, and sister to a more derived clade of subgenus Rosa sections (including Synstylae, Rosa, and Caninae) in the tree from the Bayesian analysis. Because both R. foetida and R. hugonis are members of Rosa sect. Pimpinellifoliae, this places R. persica and R. roxburghii within the subgenus Rosa, dismissing their subgeneric status. The incorporation of both subgenera in subgenus Rosa is in line with previous results on RAPDs (Jan et al., 1999; but only R. roxburghii was included), and on matK (Matsumoto et al., 1998), ITS (Wu et al., 2001b; Wissemann and Ritz, 2005), atpB-rbcL intergenic spacer (Wissemann and Ritz, 2005; but only for R. roxburghii), and trnL-F/psbA-trnH spacer sequences (Bruneau et al., 2007). The close relationship of R. persica with sect. Pimpinellifoliae is supported by the ITS data of Wissemann and Ritz (2005) and by data on pollen morphology by Ueda and Tomita (1989).
Our results indicate an early-derived position for R. persica, and a slightly more derived position for R. roxburghii. The early-derived position of R. persica is in line with the ITS data of Wu et al. (2001b) and Wissemann and Ritz (2005). The more derived position of R. roxburghii is in line with the ITS data of Wu et al. (2001b) and Wissemann and Ritz (2005), and the atpB-rbcL data of Wissemann and Ritz (2005).
Phylogenetic relationships among the sections of subgenus Rosa
Our results indicate that sect. Pimpinellifoliae may be polyphyletic, with R. spinosissima separated from the other species. However, most of the relationships involving the sect. Pimpinellifoliae species are weakly supported, and therefore the phylogenetic structure of sect. Pimpinellifoliae remains unclear. The polyphyly of sect. Pimpinellifoliae is corroborated by recent results on trnL-F/psbA-trnH spacer sequences (Bruneau et al., 2007). The separate position of R. spinosissima (especially in the MPT, Fig. 2) is in line with the view of Wissemann and Ritz (2005) that "the most divergent group within the Pimpinellifoliae are the Scotch roses themselves," and supported by data on matK sequences (Matsumoto et al., 1998).
The sect. Rosa species (R. xcentifolia, R. xdamascena, and R. gallica) and the sect. Synstylae species (R. arvensis, R. multiflora, R. wichurana, and R. moschata) are in the same clade in both phylogenies (2a/b+3a/b in Figs. 2 and 3), which could indicate a close relationship between these sections. Several of the subclades are well supported, and either contain parts of the sections (clades 2a, 3a, 3b) or mixtures thereof (clade 2b). Unfortunately, the exact relationship between the sections cannot be determined because the earlier-derived branches in the clade are only weakly supported. On top of that, accessions of the sect. Caninae species R. tomentella and R. turcica further obscure the relationships.
The sect. Synstylae is generally considered to be monophyletic, a view that is corroborated by data on RAPDs (Millan et al., 1996; Jan et al., 1999), matK (Wu et al., 2000), and ITS sequences (Matsumoto et al., 2000; Wu et al., 2001b). In our results, the section appears as a polyphyletic group including clades 3a, 3b, and a part of 2b, but the distinction between these clades is only weakly supported. Given the weak support, we consider our data to be inconclusive as to the phylogenetic structure of sect. Synstylae. With the exclusion of R. turcica 77 and R. tomentella 78 (see next paragraph), section Caninae (including subsect. Rubigineae) appears in our study as a monophyletic group (group 1 in Figs. 1–3
; note that the names on bar D in Figs. 1–3 indicate the various subsections of the monophyletic section Caninae, including the nonmonophyletic subsection Caninae). The monophyly of sect. Caninae is in line with the general consensus, and with data on RAPDs (Debener et al., 1996; Millan et al., 1996), ITS sequences (Matsumoto et al., 2000; Wissemann and Ritz, 2005), and AFLPs (Deák et al., 2005). Moreover, sect. Caninae (together with a few hybrid species that have a putative Caninae parent) is characterized by a unique type of meiosis (Täckholm, 1920; Nybom et al., 2006) and ITS sequence (Wissemann, 1999; Ritz et al., 2005).
Section Cinnamomeae is represented by 10 species in our study, forming a paraphyletic grade in the MPT (grade 4 in Fig. 2) and a paraphyletic clade in the tree from the Bayesian analysis (clade 4 in Fig. 3). The paraphyly is caused by the presence of sect. Carolinae.
The two sect. Carolinae species in our study form a monophyletic clade embedded within sect. Cinnamomeae (group 4 in Figs. 2 and 3). This position within Cinnamomeae is supported by RAPD data (Jan et al., 1999), AFLP and morphology data (Joly and Bruneau, 2007), and sequence data on ITS (Wissemann and Ritz, 2005), atpB-rbcL (Wissemann and Ritz, 2005), GAPDH Joly et al., 2006), and trnL-F/psbA-trnH (Bruneau et al., 2007). Therefore, our results corroborate the conclusion of these authors that sect. Carolinae should be included in sect. Cinnamomeae.
Phylogenetic relationships among the subsections of section Caninae
Our data set contains species from all six subsections within sect. Caninae, enabling a detailed examination of the subsectional divisions.
The species of subsect. Rubigineae are resolved as a monophyletic clade within sect. Caninae, except for R. micrantha 151 and R. turcica 77. The latter accessions are both sister to R. tomentella accessions: R. micrantha 151 is sister to R. tomentella 154 among sect. Caninae species, and R. turcica 77 is sister to R. tomentella 78 in cluster 2b outside sect. Caninae (see Figs. 2 and 3). The close connection of two individual Rubigineae accessions with R. tomentella in different parts of the trees may indicate a complex evolutionary history of R. tomentella, involving hybridization with species from subsect. Rubigineae. In light of this scenario, we consider that R. micrantha 151 and R. turcica 77 are not representative for the subsect. Rubigineae as a group.
Subsection Tomentellae is represented by two species: R. tomentella and R. abietina. Two accessions of R. tomentella are sister groups of sect. Rubigineae species, as discussed before. The third accession (R. tomentella 226) also shows close affinity to sect. Rubigineae. In the MPT (Fig. 2), it is sister to a clade with the subsect. Rubigineae species and R. stylosa 223. In the tree from the Bayesian analysis (Fig. 3), R. tomentella 226 and R. stylosa 223 form a sister group to the subsect. Rubigineae clade. Rosa abietina is sister to a polyphyletic clade with subsect. Vestitae species and R. corymbifera ‘Laxa' in the MPT (Fig. 2), and is sister to R. corymbifera ‘Laxa' in the tree from the Bayesian analysis (Fig. 3). Given these separate positions of R. tomentella and R. abietina, a subsectional status for Tomentellae does not seem warranted.
In both phylogenies, the five species from subsect. Vestitae occupy several polyphyletic clades and grades, distributed among clades and grades of species from the other subsections. The Vestitae species that are represented by multiple accessions (R. pseudoscabriuscula, R. sherardii, and R. villosa) are present in multiple clades and grades. Given this pattern, our data do not warrant a subsectional status for Vestitae. In contrast, the RAPD data of Olsson et al. (2000) showed some distinction between Vestitae (R. sherardii and R. villosa) and Caninae (R. canina and R. dumalis). However, Olsson et al. (2000) examined far fewer markers than we did in the current study, and they did not indicate the statistical support for their distinction.
The subsects. Trachyphyllae (R. jundzillii) and Rubrifoliae (R. glauca) are each represented by only one species. The sole accession of R. jundzillii is in a polyphyletic clade with subsect. Caninae species, within the range of intraspecific variation of e.g., R. dumalis and R. subcollina (Figs. 2 and 3). Therefore, our results do not support the distinction of subsect. Trachyphyllae from subsect. Caninae.
The two accessions of R. glauca group together. In the tree from the Bayesian analysis (Fig. 3), they are an early-derived sister group to a clade with the rest of the sect. Caninae species. In the MPT (Fig. 2), they are in a similar position, with only R. canina 4 as an earlier-derived accession. Therefore, our results may indicate that subsect. Rubrifoliae is an early-derived group within sect. Caninae.
Given the results described, we conclude that subsect. Rubigineae is a derived and genetically well-defined monophyletic group within sect. Caninae, while there is little support for the recognition of the remaining subsections. The distinct position of subsect. Rubigineae is supported by earlier results on RAPDs (Olsson et al., 2000) and morphology (discussed in Wissemann, 2000). The recognition of subsect. Rubigineae as a monophyletic group within sect. Caninae renders the remaining part of the section paraphyletic. Therefore, the acceptance of subsect. Rubigineae as a separate taxonomic unit depends on one's view on the acceptability of paraphyletic taxa. If only monophyletic taxa are to be recognized, the underlying assumption is that the development of new groups coincides with the extinction of the ancestral group(s). As was pointed out by Sosef (1997) and Hörandl (2006), for example, actual evolution may be more complex than that, and new groups may develop alongside extant ancestors. The latter seems to have been the case in sect. Caninae; therefore, we consider that the recognition of subsect. Rubigineae as a separate taxon within sect. Caninae does the most justice to the evolutionary history of these groups. A more elaborate discussion of paraphyletic vs. monophyletic groups is beyond the scope of the present paper, and the reader is referred to Mayr and Bock (2002) and Hörandl (2006), and references therein.
The application of AFLP data to infer phylogenies of complex groups
In the current study, we used AFLP markers as a source of phylogenetic information. In contrast to RFLPs and the various sequences mentioned in the introduction, AFLPs provided sufficient variation for reconstructing well-resolved trees. The present results are generally congruent with previous morphology-based classifications (discussed in Wissemann and Ritz, 2005) and with results based on other molecular markers (discussed in the previous sections). Moreover, the AFLP based trees have many well-supported branches, especially on the accession level. However, the AFLP trees also have many branches with poor support, especially on the higher taxonomic levels.
In typical AFLP data sets, the markers represent an anonymous sample across (more or less) the entire genome. This sampling is usually considered to be an advantage, because it ensures the retrieval of species trees (with terminals representing the entire species) rather than gene-trees (with genes on the terminals, whose position may not be representative for that of the species). However, in complex groups such as Rosa, species are present for which different parts of the genome have different evolutionary histories, as in sect. Caninae (Wissemann, 1999, Wissemann, 2002; Ritz et al., 2005; Nybom et al., 2006). Most of these histories will be represented in the genome-wide AFLP sample, which will therefore contain conflicting signal. As a result, the support values will be low for those parts of the tree containing the species involved. Nevertheless, even the poorly supported parts of the trees show a general congruence with classifications based on morphological and molecular data, indicating that the conflict did not completely obscure the signal. Given this congruence for poorly supported clades and the presence of support for other clades in the trees, we conclude that AFLPs can be a valuable source of phylogenetic information, even in species groups with a complex evolutionary history such as Rosa.
Bayesian inference on AFLP data
According to Ronquist et al. (2005), there are two major criteria to establish the burn-in value for the calculation of the posterior probabilities. Firstly, the log likelihood values of the cold chain should have reached a stable value, and secondly, the two independent runs should have converged. The target value for convergence is an average standard deviation of split frequencies of 0.01 (Ronquist et al., 2005). In our study, the log likelihood values of both runs were oscillating around roughly the same value (see online Appendix S2). This similarity in value indicates that both runs converged on the same solution and hence that the analysis had reached a stationary phase that warrants a reliable sampling of trees for the calculation of posterior probability values. However, there is also a considerable amount of variation around the value the runs converged on (Appendix S2). This variation may be caused by the presence of conflicting signal in the data set, which prevents the analysis from reaching a more stable (and less variable) solution. Such conflict may be caused by the complex evolutionary history of the genus Rosa, as discussed earlier. The conflict in signal may also be the reason for the slow convergence of both runs, as indicated by online appendix S3. If this explanation is correct, a target convergence value of 0.01 may be unfeasible for the present analysis, as well as for other analyses involving species groups with a complex evolutionary history.
Considering the branch supports, our results show that the posterior probabilities in the tree from the Bayesian analysis are generally higher than the bootstrap support values for the corresponding branches in the MPT. This finding is in line with previous studies (Wilcox et al., 2002; Erixon et al., 2003, and references therein), as was our finding that posterior probabilities and bootstrap support values are not tightly correlated. In a study of Whittingham et al. (2002), all branches with >0.95 posterior probability received a maximum likelihood bootstrap support 
70%, and a branch probability of 0.80 corresponded to a bootstrap value of 
70%. However, Whittingham et al. (2002) stated that their results did not allow for the establishment of a general rule of thumb because there was a substantial amount of variation around these values. A study of Leaché and Reeder (2002) also showed that posterior probabilities and bootstrap support values were not tightly correlated, with parsimony bootstrap values being less tightly correlated to posterior probabilities than maximum likelihood bootstrap values. This lack of a tight correlation may be explained by the fact that posterior probabilities and (parsimony) bootstrap support values measure different characteristics of the data (Alfaro et al., 2003). Posterior probabilities measure the probability of a node being correct, conditional on the data set at hand and the evolutionary model employed. As such, their reliability heavily depends on the accuracy of both the data sample and the evolutionary model. Parsimony bootstrap values are different in that they measure the sensitivity of the observed results to the sampling error associated with collecting characters from a hypothesized underlying character distribution, without the use of an explicitly specified evolutionary model. Given the fact that posterior probabilities and parsimony bootstrapping measure different features of the data, the most reliable nodes are those that have high values of both. Figures 2 and 3 show that such nodes are usually connecting terminal taxa, which corroborates the general notion that AFLP markers are most reliable at the lower taxonomic levels (Mueller and Wolfenbarger, 1999).
Parsimony vs. Bayesian inference
We used two methods of phylogeny reconstruction: parsimony analysis and Bayesian inference. A comparison of the results (Figs. 2 and 3) shows that both trees are very similar, especially for the well-supported branches. For each of the methods, our data set violated important assumptions: the parsimony assumption was violated by the complex and reticulate evolutionary history of the genus Rosa, and the restriction site model in MrBayes was violated by the fact that the evolution of AFLP markers is far more complex than the model assumes (Luo et al., 2007). Nevertheless, both trees are generally congruent with morphology-based classifications and with previous molecular phylogenies, which makes the results appear to be reliable. Apparently, neither of the violations was severe enough to seriously hamper the analyses. Interestingly, this may imply that the F81-like restriction site model employed in MrBayes, which only takes into account the band frequencies, is specific enough to adequately model the evolution of AFLP fragments for purposes of Bayesian inference.
The most striking difference between the parsimony and Bayesian analyses was the amount of time involved. The parsimony analysis performed on a PowerMac G4 800Mhz was finished in 12 h, with a few additional seconds for the branch swapping, and 20 h for the bootstrapping. The Bayesian inference performed on a 3 GHz Windows PC took 430 hours, which is more than 13-fold higher. Although 430 h is a substantial amount of time, it is not prohibitively long. Given the fact that the resulting tree was very similar to that of the parsimony analysis, we conclude that Bayesian inference as implemented in MrBayes is a useful alternative to parsimony analysis of AFLP data.
Conclusions
We used a data set of Rosa species to examine the applicability of AFLP markers for relationship studies in species groups with a complicated and reticulate evolutionary history. In our analyses, AFLPs yielded sufficient variation to reconstruct well-resolved trees, but many of the branches were poorly supported. Nevertheless, a general congruence with morphology-based classifications and with previous studies based on molecular markers and sequences indicated that the reconstructed relationships were reliable.
To infer the Rosa phylogenies, we used both Wagner parsimony and Bayesian inference. Although the parsimony assumption is violated by the complex and reticulate evolution in the genus Rosa, the presence of well-supported branches in the tree and a general congruence with previous morphological and molecular studies indicated that the tree topologies are reliable. For the Bayesian inference, the assumptions of the restriction site model are violated by the complex evolution of AFLP markers. However, the presence of well supported branches in the tree from the Bayesian analysis, and a general congruence with the MPT indicated that the Bayesian inference yielded a reliable phylogeny as well. Therefore, we conclude that Bayesian inference is a useful alternative to parsimony analysis of AFLP data. The most important drawback of Bayesian inference is that it is more time consuming than parsimony analysis.
Phenetic analysis of the AFLP data revealed two major clusters of cultivars, those related to sect. Rosa (species R.xdamascena and R. gallica) or to sect. Synstylae (wild species R. moschata, R. wichurana, and R. multiflora), respectively. Phylogenetic analysis of the AFLP data yielded several conclusions on species relationships. (1) Subgenera Hulthemia and Platyrhodon do not deserve a subgeneric status, but must be included in subgenus Rosa. Subgenus Hulthemia is an early-derived group, subgenus Platyrhodon is more derived. (2) Section Carolinae should be included in sect. Cinnamomeae. (3) Section Caninae including subsection Rubigineae is a monophyletic group. (4) Within sect. Caninae, subsect. Rubigineae is a monophyletic group. Our data do not support the remaining subsections in Caninae. (5) The recognition of subsect. Rubigineae as a separate taxon will render the remaining sect. Caninae paraphyletic, and therefore depends on one's willingness to accept paraphyletic taxa. Given the complex and reticulate evolutionary patterns in sect. Caninae and other Rosa groups, the lack of a single hierarchy may make the recognition of such a paraphyletic subsect. Rubigineae acceptable.
FOOTNOTES
1 The authors thank Prof. Dr. M.S.M. Sosef (Nationaal Herbarium Nederland-Wageningen branch, Biosystematics Group, Wageningen University, The Netherlands) for discussing the merits of paraphyletic groups, Mrs. M. Lemmens-Pott and Mr. H. de Leeuw (Nationaal Herbarium Nederland-Wageningen branch, Biosystematics Group, Wageningen University, The Netherlands) for information and discussion on Rosa provenances from botanical gardens De Dreijen and Belmonte, and Dr. L. Leus (Unit Plant, Applied Genetics and Breeding, Institute for Agricultural and Fisheries Research, Melle, Belgium) for collecting herbarium material at Rozentuin Mechelen. This study has been carried out with financial support from the Netherlands Ministry of Agriculture, Nature and Food Safety, and from the Commission of the European Community (QLRT-2001-01278, Genetic evaluation of European rose resources for conservation and horticultural use [Generose]). This study does not necessarily reflect the Commission's views and in no way anticipates the Commission's future policy in this area. 
9 Current address: Koopman Scientific Services, P.O. Box 404, 1600 AK Enkhuizen, The Netherlands
10 Author for correspondence (e-mail: koopman{at}koopmanscientific.com)
LITERATURE CITED
Alfaro, M. E., S. Zoller, AND F. Lutzoni. 2003. Bayes or bootstrap? A simulation study comparing the performance of Bayesian Markov chain Monte Carlo sampling and bootstrapping in assessing phylogenetic confidence. Molecular Biology and Evolution 20: 255–266.
Arens, P., H. Coops, J. Jansen, AND B. Vosman. 1998. Molecular genetic analysis of black poplar (Populus nigra L.) along Dutch rivers. Molecular Ecology 7: 11–18.
Bagley, M. J., S. L. Anderson, AND B. May. 2001. Choice of methodology for assessing genetic impacts of environmental stressors: polymorphism and reproducibility of RAPD and AFLP fingerprints. Ecotoxicology (London, England) 10: 239–244.
Brouat, C., D. Mckey, AND E. J. P. Douzery. 2004. Differentiation in a geographical mosaic of plants coevolving with ants: phylogeny of the Leonardoxa africana complex (Fabaceae: Caesalpinioideae) using amplified fragment length polymorphism markers. Molecular Ecology 13: 1157–1171.[CrossRef][Medline]
Bruneau, A., S. Joly, J. R. Starr, AND J.-N. Drouin. 2005. Molecular markers indicate that the narrow Québec endemics Rosa rousseauiorum and Rosa williamsii are synonymous with the widespread Rosa blanda. Canadian Journal of Botany 83: 386–398.[CrossRef]
Bruneau, A., J. R. Starr, AND S. Joly. 2007. Phylogenetic relationships in the genus Rosa: new evidence from chloroplast DNA sequences and an appraisal of current knowledge. Systematic Botany 32: 366–378.[CrossRef][Web of Science]
Cregan, P. B. 1992. Simple sequence repeat DNA length polymorphisms. Probe 2: 18–22.
Deák, T., G. Facsar, AND G. D. Bisztray. 2005. Genetic polymorphism of Rosa genotypes native to Hungary. Acta Horticulturae 690: 57–62.
Debener, T., C. Bartels, AND L. Mattiesch. 1996. RAPD analysis of genetic variation between a group of rose cultivars and selected wild rose species. Molecular Breeding 2: 321–327.[CrossRef][Web of Science]
Eriksson, T., M. S. Hibbs, A. D. Yoder, C. F. Delwiche, AND M. J. Donoghue. 2003. The phylogeny of Rosoideae (Rosaceae) based on sequences of the internal transcribed spacers (ITS) of nuclear ribosomal DNA and the trnL/F region of chloroplast DNA. International Journal of Plant Sciences 164: 197–211.[CrossRef][Web of Science]
Erixon, P., B. Svennblad, T. Britton, AND B. Oxelman. 2003. Reliability of Bayesian posterior probabilities and bootstrap frequencies in phylogenetics. Systematic Biology 52: 665–673.
Esselink, G. D., M. J. M. Smulders, AND B. Vosman. 2003. Identification of cut rose (Rosa hybrida) and rootstock varieties using robust sequence tagged microsatellite site markers. Theoretical and Applied Genetics 106: 277–286.[Web of Science][Medline]
Farris, J. S. 1989. The retention index and the rescaled consistency index. Cladistics 5: 417–419.[CrossRef][Web of Science]
Felsenstein, J. 1981. Evolutionary trees from DNA sequences: a maximum likelihood approach. Journal of Molecular Evolution 17: 368–376.[CrossRef][Web of Science][Medline]
Felsenstein, J. 1985. Confidence limits on phylogenies: an approach using the bootstrap. Evolution; International Journal of Organic Evolution 39: 783–791.
Hillis, D. M., AND J. P. Huelsenbeck. 1992. Signal, noise, and reliability in molecular phylogenetic analyses. Journal of Heredity 83: 189–195.
Hillis, D. M., C. Moritz, AND B. K. Mable. 1996. Applications of molecular systematics. In D. M. Hillis, C. Moritz, and B. K. Mable [eds.], Molecular Systematics, 515–543. Sinauer, Sunderland, Massachusetts, USA.
Hörandl, E. 2006. Paraphyletic versus monophyletic taxa-evolutionary versus cladistic classifications. Taxon 55: 564–570.[Web of Science]
Huelsenbeck, J. P., AND F. Ronquist. 2001. MrBayes: Bayesian inference of phylogenetic trees. Bioinformatics (Oxford, England) 17: 754–755.[CrossRef]
Jan, C. H., D. H. Byrne, J. Manhart, AND H. Wilson. 1999. Rose germplasm analysis with RAPD markers. HortScience 34: 341–345.
Joly, S., AND A. Bruneau. 2006. Incorporating allelic variation for reconstructing the evolutionary history of organisms from multiple genes: An example from Rosa in North America. Systematic Biology 55: 623–636.[CrossRef][Medline]
Joly, S., AND A. Bruneau. 2007. Delimiting species boundaries in Rosa sect. Cinnamomeae (Rosaceae) in eastern North America. Systematic Botany 32: 819–836.[Web of Science]
Joly, S., J. R. Starr, W. H. Lewis, AND A. Bruneau. 2006. Polyploid and hybrid evolution in roses east of the Rocky Mountains. American Journal of Botany 93: 412–425.
Jones, C. J., K. J. Edwards, S. Castaglione, M. O. Winfield, F. Sala, C. van de Wiel, G. Bredemeijer, B. Vosman, M. Matthes, A. Daly, R. Brettschneider, P. Bettini, M. Buiatti, E. Maestri, A. Malcevschi, N. Marmiroli, R. Aert, G. Volckaert, J. Rueda, R. Linacero, A. Vazquez, AND A. Karp. 1997. Reproducibility testing of RAPD, AFLP and SSR markers in plants by a network of European laboratories. Molecular Breeding 3: 381–390.[CrossRef][Web of Science]
Kardolus, J. P., H. J. Van Eck, AND R. G. Van den Berg. 1998. The potential of AFLPs in biosystematics: A first application in Solanum taxonomy (Solanaceae). Plant Systematics and Evolution 210: 87–103.[CrossRef][Web of Science]
Kluge, A. G., AND J. S. Farris. 1969. Quantitative phyletics and the evolution of anurans. Systematic Zoology 18: 1–32.
Koopman, W. J. M. 2005. Phylogenetic signal in AFLP data sets. Systematic Biology 54: 197–217.
Koopman, W. J. M., M. J. Zevenbergen, AND R. G. Van den Berg. 2001. Species relationships in Lactuca s.l. (Lactuceae, Asteraceae) inferred from AFLP fingerprints. American Journal of Botany 88: 1881–1887.
Leaché, A. D., AND T. W. Reeder. 2002. Molecular systematics of the eastern fence lizard (Sceloporus undulatus): a comparison of parsimony, likelihood, and Bayesian approaches. Systematic Biology 51: 44–68.
Leus, L., F. Jeanneteau, J. van Huylenbroeck, E. van Bockstaele, AND J. de Riek. 2004. Molecular evaluation of a collection of rose species and cultivars by AFLP, ITS, rbc L, and mat K. Acta Horticulturae 651: 141–147.
Lewis, P. O., M. T. Holder, AND K. E. Holsinger. 2005. Polytomies and Bayesian phylogenetic inference. Systematic Biology 54: 241–253.
Luo, R., A. L. Hipp, AND B. Larget. 2007. A Bayesian model of AFLP marker evolution and phylogenetic inference. Statistical Applications in Genetics and Molecular Biology (1): article 11.
Mace, E. S., C. G. Gebhardt, AND R. N. Lester. 1999a. AFLP analysis of genetic relationships in the tribe Datureae (Solanaceae). Theoretical and Applied Genetics 99: 634–641.[CrossRef][Web of Science]
Mace, E. S., R. N. Lester, AND C. G. Gebhardt. 1999b. AFLP analysis of genetic relationships among the cultivated eggplant, Solanum melongena L., and wild relatives (Solanaceae). Theoretical and Applied Genetics 99: 626–633.[CrossRef][Web of Science]
Matsumoto, S., M. Kouchi, H. Fukui, AND Y. Ueda. 2000. Phylogenetic analyses of the subgenus Eurosa using the ITS nrDNA sequence. Acta Horticulturae 521: 193–202.
Matsumoto, S., M. Kouchi, J. Yabuki, M. Kusunoki, Y. Ueda, AND H. Fukui. 1998. Phylogenetic analyses of the genus Rosa using the mat K sequence: molecular evidence for the narrow genetic background of modern roses. Scientia Horticulturae 77: 73–82.[CrossRef]
Matsumoto, S., H. Wakita, AND H. Fukui. 1997. Molecular classification of wild roses using organelle DNA probes. Scientia Horticulturae 68: 191–196.[CrossRef]
Mayr, E., AND W. J. Bock. 2002. Classifications and other ordering systems. Journal of Zoological Systematics and Evolutionary Research 40: 169–194.[CrossRef][Web of Science]
Millan, T., F. Osuna, S. Cobos, A. M. Torres, AND J. I. Cubero. 1996. Using RAPDs to study phylogenetic relationships in Rosa. Theoretical and Applied Genetics 92: 273–277.[CrossRef][Web of Science]
Mueller, U. G., AND L. L. Wolfenbarger. 1999. AFLP genotyping and fingerprinting. Trends in Ecology &Evolution 14: 389–394.[CrossRef][Web of Science][Medline]
Nei, M., AND W.-H. Li. 1979. Mathematical model for studying genetic variation in terms of restriction endonucleases. Proceedings of the National Academy of Sciences, USA 76: 5269–5273.
Nybom, H. 2004. Comparison of different nuclear DNA markers for estimating intraspecific genetic diversity in plants. Molecular Ecology 13: 1143–1155.[CrossRef][Medline]
Nybom, H., G. D. Esselink, G. Werlemark, L. Leus, AND B. Vosman. 2006. Unique genomic configuration revealed by microsatellite DNA in polymorphic dogroses, Rosa sect. Caninae. Journal of Evolutionary Biology 19: 635–648.[CrossRef][Medline]
Olsson, Å., H. Nybom, AND H. C. Prentice. 2000. Relationships between Nordic dogroses (Rosa L. sect. Caninae, Rosaceae) assessed by RAPDs and elliptic fourier analysis of leaflet shape. Systematic Botany 25: 511–521.[CrossRef][Web of Science]
Page, R. D. M., AND E. C. Holmes. 1998. Molecular evolution: A phylogenetic approach. Blackwell Science, Oxford, UK.
Rehder, A. 1940. Manual of cultivated trees and shrubs. MacMillan, New York, New York, USA.
Ritz, C. M., H. Schmuths, AND V. Wissemann. 2005. Evolution by reticulation: European dogroses originated by multiple hybridization across the genus Rosa. Journal of Heredity 96: 4–14.
Rohlf, F. J. 1993. NTSYS-pc, numerical taxonomy and multivariate analysis system. Applied Biostatistics, Setauket, New York, USA.
Ronquist, F., AND J. P. Huelsenbeck. 2003. MrBayes 3: Bayesian phylogenetic inference under mixed models. Bioinformatics (Oxford, England) 19: 1572–1574.[CrossRef]
Ronquist, F., J. P. Huelsenbeck, AND P. Van der Mark. 2005. MrBayes 3. 1 Manual, draft 5/26/2005. School of Computational Science, Florida State University, Tallahassee, Florida, USA.
Saitou, N., AND M. Nei. 1987. The neighbor-joining method: A new method for reconstructing phylogenetic trees. Molecular Biology and Evolution 4: 406–425.[Abstract]
Scariot, V., A. Akkak, AND R. Botta. 2006. Characterization and genetic relationships of wild species and old garden roses based on microsatellite analysis. Journal of the American Society for Horticultural Science 131: 66–73.
Smulders, M. J. M., J. van der Schoot, R. H. E. M. Geerts, A. G. Antonisse-de Jong, H. Korevaar, A. van der Werf, AND B. Vosman. 2000. Genetic diversity and the reintroduction of meadow species. Plant Biology 2: 447–454.[CrossRef]
Sosef, M. S. M. 1997. Hierarchical models, reticulate evolution and the inevitability of paraphyletic supraspecific taxa. Taxon 46: 75–85.[CrossRef][Web of Science]
Swofford, D. L. 1999. PAUP*: Phylogenetic analysis using parsimony (*and other methods), version 4.0b10 (Altivec). Sinauer, Sunderland, Massachusetts, USA.
Swofford, D. L., AND D. P. Begle. 1993. PAUP: Phylogenetic analysis using parsimony, version 3.1. User's manual. Laboratory of Molecular Systematics, Washington D.
Täckholm, G. 1920. On the cytology of the genus Rosa. A preliminary note. Svensk Botanisk Tidskrift 14: 300–311.
Takeuchi, S., K. Nomura, H. Uchiyama, AND K. Yoneda. 2000. Phylogenetic relationship in the genus Rosa based on the restriction enzyme analysis of the chloroplast DNA. Journal of the Japanese Society for Horticultural Science 69: 598–604.[CrossRef][Web of Science]
Ueda, Y., AND H. Tomita. 1989. Morphometric analysis of pollen exine patterns in roses. Journal of the Japanese Society for Horticultural Science 58: 211–220.[CrossRef][Web of Science]
Van de Peer, Y., AND R. De Wachter. 1994. TREECON for Windows: A software package for the construction and drawing of evolutionary trees for the Microsoft Windows environment. Computer Applications in the Biosciences 10: 569–570.
Vos, P., R. Hogers, M. Bleeker, M. Reijans, T. van de Lee, M. Hornes, A. Frijters, J. Pot, J. Peleman, M. Kuiper, AND M. Zabeau. 1995. AFLP: a new technique for DNA fingerprinting. Nucleic Acids Research 23: 4407–4414.
Vriesendorp, B. 2007. Phylogenetworks: Exploring reticulate evolution and its consequences for phylogenetic reconstruction. Ph.D. dissertation, Wageningen University, Wageningen, Netherlands.
Vriesendorp, B., AND F. T. Bakker. 2005. Reconstructing patterns of reticulate evolution in angiosperms: what can we do? Taxon 54: 593–604.[Web of Science]
Wen, X. P., X. M. Pang, AND X. X. Deng. 2004. Characterization of genetic relationships of Rosa roxburghii Tratt and its relatives using morphological traits, RAPD and AFLP markers. Journal of Horticultural Science &Biotechnology 79: 189–196.[Web of Science]
Whittingham, L. A., B. Slikas, D. W. Winkler, AND F. H. Sheldon. 2002. Phylogeny of the tree swallow genus, Tachycineta (Aves: Hirundinidae), by Bayesian analysis of mitochondrial DNA sequences. Molecular Phylogenetics and Evolution 22: 430–441.[CrossRef][Web of Science][Medline]
Wilcox, T. P., D. J. Zwickl, T. A. Heath, AND D. M. Hillis. 2002. Phylogenetic relationships of the dwarf boas and a comparison of Bayesian and bootstrap measures of phylogenetic support. Molecular Phylogenetics and Evolution 25: 361–371.[CrossRef][Web of Science][Medline]
Wissemann, V. 1999. Genetic constitution of Rosa sect. CaninaeR. canina, R. jundzillii) and sect. GallicanaeR. gallica). Journal of Applied Botany 73: 191–196.
Wissemann, V. 2000. Epicuticular wax morphology and the taxonomy of Rosa (section Caninae, subsection Rubiginosae). Plant Systematics and Evolution 221: 107–112.[CrossRef][Web of Science]
Wissemann, V. 2002. Molecular evidence for allopolyploid origin of the Rosa canina-complex (Rosaceae, Rosoideae). Journal of Applied Botany 76: 176–178.
Wissemann, V. 2003. Conventional taxonomy of wild roses. In A. Roberts, T. Debener, and S. Gudin [eds.], Encyclopedia of rose science, 111–117. Elsevier, London, UK.
Wissemann, V., AND C. M. Ritz. 2005. The genus Rosa (Rosoideae, Rosaceae) revisited: molecular analysis of nrITS-1 and atp B-rbc L intergenic spacer (IGS) versus conventional taxonomy. Botanical Journal of the Linnean Society 147: 275–290.[CrossRef][Web of Science]
Wu, S., S. Nishihara, AND Y. Ueda. 2001a. Phylogenetic analysis of section Synstylae in the genus Rosa based on RAPD markers. Acta Horticulturae 547: 391–402.
Wu, S. Q., Y. Ueda, H. He, S. Nishihara, AND S. Matsumoto. 2000. Phylogenetic analysis of Japanese Rosa species using mat K sequences. Breeding Science 50: 275–281.[Web of Science]
Wu, S., Y. Ueda, S. Nishihara, AND S. Matsumoto. 2001b. Phylogenetic analysis of Japanese Rosa species using DNA sequences of nuclear ribosomal internal transcribed spacers (ITS). The Journal of Horticultural Science &Biotechnology 76: 127–132.
Wylie, A. P. 1954. The history of garden roses, part I. Journal of the Royal Horticultural Society, London 79: 555–571.
Yan, Z., C. Denneboom, A. Hattendorf, O. Dolstra, T. Debener, P. Stam, AND P. B. Visser. 2005. Construction of an integrated map of rose with AFLP, SSR, PK, RGA, RFLP, SCAR and morphological markers. Theoretical and Applied Genetics 110: 766–777.[CrossRef][Medline]
Yang, Z., AND B. Rannala. 1997. Bayesian phylogenetic inference using DNA sequences: a Markov chain Monte Carlo method. Molecular Biology and Evolution 14: 717–724.[Abstract]
CiteULike Complore Connotea Del.icio.us Digg Facebook Reddit Technorati Twitter What's this?
This article has been cited by other articles:
|
|
 |
|
  W. K. Savage and S. P. Mullen A single origin of Batesian mimicry among hybridizing populations of admiral butterflies (Limenitis arthemis) rejects an evolutionary reversion to the ancestral phenotype Proc R Soc B, July 22, 2009; 276(1667): 2557 - 2565. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 K. De Cock, K. Vander Mijnsbrugge, P. Breyne, E. Van Bockstaele, and J. Van Slycken Morphological and AFLP-based Differentiation within the Taxonomical Complex Section Caninae (subgenus Rosa) Ann. Bot., November 1, 2008; 102(5): 685 - 697. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
