Nuclear and chloroplast DNA-based phylogenies of Chrysanthemoides Tourn. ex Medik. (Calenduleae; Asteraceae) reveal extensive incongruence and generic paraphyly, but support the recognition of infraspecific taxa in C. monilifera

The small genus Chrysanthemoides comprises two species within which a number of infraspecific taxa have been recognized, some of which are invasive aliens in Australia and New Zealand. Here we investigate the relationships of the species and infraspecific taxa using both chloroplast and nuclear non-coding DNA sequence data. Results of the analyses of the plastid and nuclear data sets are incongruent, and neither Chrysanthemoides nor Osteospermum is resolved as monophyletic, although there is some support for the recognition of infraspecific taxa. Analyses of the separate and combined data sets resolve two clades within Chrysanthemoides (which include some species of Osteospermum), and these appear to have a geographic basis, one being restricted to the mainly winter rainfall region, the other the eastern bi-seasonal rainfall area. Our results suggest that there is evidence of past or ongoing hybridization within and possibly between these two lineages. © 2009 SAAB. Published by Elsevier B.V. All rights reserved.


Introduction
The tribe Calenduleae comprises some 120 species and 12 genera (Nordenstam, 2006(Nordenstam, , 2007Nordenstam and Källersjö, 2009). Earlier assessments of the tribe recognized seven (Norlindh, 1977) or ten genera (Nordenstam, 1994). Since 1994 there have been several taxonomic re-arrangements. Nordenstam (1994Nordenstam ( , 1996 considered the previously recognized genus Castalis Cass. and sect. Blaxium (Cass.) T. Norl. of Osteospermum L. to belong to Dimorphotheca Vaill. and revived the genera Oligocarpus Less. and Tripteris Less., both of which had been included in Osteospermum by Norlindh (1943Norlindh ( , 1977. Nordenstam (1996) also questioned the monophyly of the reduced Osteospermum, noting that despite these taxonomic changes, it was still heterogeneous. Further splits from Osteospermum were described as the new genera Norlindhia B. Nord. and Monoculus B. Nord. (Nordenstam, 2006), and a species of Gibbaria Cass. was recognized as a distinct new genus Nephrotheca B. Nord. & Källersjö . This move left Gibbaria as a monotypic genus until a second species was recently added by a transfer from Osteospermum (Nordenstam and Källersjö, 2009). Norlindh (1943) established the genus Chrysanthemoides Tourn. ex Medik. for two species of Osteospermum that had fleshy drupe-like fruits. However, fleshy or semi-drupaceous fruits have later also been reported in some species of Osteospermum such as O. junceum Berg., O. asperulum (DC.) T. Norl., O. corymbosum L. and O. triquetrum L. f. (Wood and Nordenstam, 2003). The evolution of drupes or fruits with a fleshy exocarp is extremely rare Table 1 Comparison of the species and infraspecific taxon names recognized by Norlindh (1943) and the unpublished entities recognized by Griffioen (1995), with key characteristics of each entity provided. Norlindh (1943) Griffioen (1995) Morphological characteristics Habit Leaves Involucral scales Drupes Ecology C. monilifera subsp. monilifera "C. monilifera subsp. monilifera" Shrub, 1.5 m in height × 2.0 m in diameter, stems tanniferous, pubescence absent.
Found in disturbed sites on margins of climax vegetation in a range of soils including TMS, limestone and loams in the SW Cape, on mountain slopes from Piketberg in the north to Worcester and Hermanus in the east. "C. monilifera subsp. floribunda form 1" Small to large erect bush, 1-3.5 m in height× 1-6 m diameter, stems tanniferous, pubescence absent from young tissues.
Coastal sand dunes, limestones and along roadsides from Langebaan to Knysna.
Broadly ovate, covered with a loose pubescence.
Costal dunes from Knysna to Hiumansdorp, spreading into forest edges in Knysna and George area.
Grahamstown eastwards into Transkei region, in fynbos or false fynbos, shallow but fertile soils at altitudes between 1000 and 1500 m a.s.l. "C. monilifera subsp. pisifera var. pisifera form 2" Small shrub to 1.5 m high. in the Asteraceae, and as far as is known, no other Old World members of this family have drupes (Norlindh, 1977). Norlindh (1943) recognized two species in Chrysanthemoides, viz. C. incana (Burm. f.) T. Norl. and C. monilifera (L.) T. Norl., and further divided the latter into five subspecies, some of which were noted to be highly variable. In an unpublished thesis, Griffioen (1995) used morphological data supplemented with isozyme and ecological data to re-assess the infraspecific taxonomy of both species in the genus. A phenetic analysis of morphological characters indicated that C. incana could be distinguished from C. monilifera on the basis of spinescence, prostrate growth form, and the distribution of pubescence on the stems, leaves and receptacles in the former species (Griffioen, 1995). These differences are accompanied by various geographic and ecological attributes. Furthermore, isozyme electrophoresis indicated that the two species are distinct, and do not hybridize when co-occurring. Within C. incana, Griffioen (1995) recognized six infraspecific taxa, and within C. monilifera 10 taxa were recognized (at subspecies, variety and form rank; Table 1). Norlindh (1943) noted that the drupes of Chrysanthemoides are edible and most likely bird dispersed, and he ascribed the presence of the species on St. Helena since before 1839 to bird dispersal. Chrysanthemoides fruits are eaten by a variety of birds in South Africa (Rowan, 1967;Keath et al., 1992;Joffe, 2001). In St. Helena dispersal by the introduced Indian Myna has been reported (Ashmole and Ashmole, 2000). In Australia, where Chrysanthemoides is an invasive weed, various mammals have been recorded as dispersers as well as emus and flying frugivorous birds (Weiss, 1986;Meek, 1998). Man has thus also acted as a dispersal agent, and both C. monilifera subsp. rotundata (DC.) T. Norl. and C. monilifera subsp. monilifera are legislated as "weeds of national significance" in Australia (http://www.weeds.org.au/WoNS/bitoubush/), and there has been over two decades of biocontrol research in Australia on C. monilifera (Downie et al., 2007). This species is also a problem in New Zealand (Roy et al., 2004). The study of these taxa in their native region is thus of vital importance if the spread of these species as weeds is to be controlled (Scott, 1996). However, the fact that these taxa are able to establish easily has resulted in their use in rehabilitation efforts following mining activities (Hälbich, 2003) and to stabilize coastal dunes in urban areas of South Africa (Nichols, 1996).
Chrysanthemoides is distributed across a number of biomes and vegetation types in southern Africa, ranging from the Fynbos of the South Western Cape, to the montane Grassland of the Drakensberg, Chimanimani and mountains of eastern Africa (Griffioen, 1995). Of the two recognized species, C. incana is mostly restricted to the South Western Cape but extends mainly along the coast northwards to Namaqualand and southern Namibia (Angra Pequena) and along the southern coast to about Cape Agulhas. C. monilifera is more widespread, with C. monilifera subsp. canescens (DC.) T. Norl. extending from the Eastern Cape mountains and the Drakensberg into northern Transvaal, and C. monilifera subsp. septentrionalis T. Norl. distributed in the montane regions of Tropical East Africa from Zimbabwe north to Tanzania (Norlindh, 1943).
Thus, while Chrysanthemoides in current taxonomy comprises only two species, there is ample evidence (published and unpublished) of considerable variation within each species. Here we use both nuclear and chloroplast DNA sequence data to undertake a phylogenetic analysis of the genus to test not only the monophyly of the genus, but also to assess if there is any genetic evidence to support the recognition of the infraspecific taxa as recognized by either or both Norlindh (1943) and Griffioen (1995), especially within C. monilifera. We emphasise that while we are using the taxa as recognized by Griffioen, they are not validly published, and hence the names in Table 1 and the text that follows appear in quotation marks to indicate this status.
In order to test taxon monophyly, we adopt a multiple exemplar approach, including two or more specimens representative of each taxonomic entity. This approach is important, as some molecular studies have indicated that species nonmonophyly (as determined by molecular data) can be quite common (e.g. Crisp and Chandler, 1996;Ohsako and Ohnishi, 2000;Syring et al., 2007;Howis et al., in press;Ramdhani et al., in press), and when noted requires careful analysis and explanation. Monophyly at infraspecific ranks is likely to be compromised by hybridization and incomplete lineage sorting. However, as noted by Holder et al. (2001), distinguishing between these two processes is difficult, but a phylogeographic approach may enable us to identify monophyletic infraspecific taxa. If so, then we suggest that this be viewed as evidence favouring their recognition as valid infraspecific taxa, and certainly as "Evolutionary Significant Units" (ESUs sensu Ryder, 1986;cf. Fraser and Bernatchez, 2001).
Two widely utilised chloroplast spacer regions were selected for this investigation: the psbA-trnH spacer, and the trnL intron in conjunction with the associated trnL-trnF spacer (hereafter termed the trnL-F region). The psbA-trnH region is being increasingly used in phylogenetic studies at the intrageneric level (Gielly et al., 1996;Sang et al., 1997;Kim et al., 1999;Chandler et al., 2001;Pelser et al., 2003;McKenzie et al., 2006;McKenzie and Barker, 2008) as well as the intraspecific level (Štorchová and Olson, 2004;Yamashiro et al., 2004). The trnL-trnF intergenic spacer region is one of the most commonly used non-coding regions of cpDNA in phylogenetic studies at the intrageneric and species level, and has occasionally been found to be sufficiently variable for use below the species level (Barker et al., 2005).
While the use of the Internal Transcribed Spacer (ITS) region for phylogenetic purposes is considered controversial by some (see for example Alvarez and Wendel, 2003;Bailey et al., 2003;Small et al., 2004 for critiques), it is still the most tractable nuclear region for molecular systematics at the species and genus level (e.g. Feliner and Rosselló, 2007;Mort et al., 2007), and has been widely used in phylogenetics of many groups of Asteraceae. Baldwin (1993) was probably the first to notice intraspecific variability in ITS sequences in the Asteraceae, and this region has been used in phylogeographic studies in the Asteraceae (for example Comes and Abbott 2001;Simurda et al., 2005;Zachariades et al., 2004;Pelser et al., 2007). However, at least some of these studies use ITS data in conjunction with additional data such as AFLP's or plastid DNA data so as to address the potentially problematic issues or paralogy and lineage sorting.

Sampling, DNA extraction, amplification and sequencing
Multiple samples from each of the species of Chrysanthemoides were obtained, and in particular we focused on ensuring coverage of the various subspecific entities within C. monilifera as recognized by Norlindh (1943) and Griffioen (1995). Where possible, we collected material at sites that had been sampled for Griffioen's phenetic analysis (Griffioen, 1995). All samples were identified to subspecies, variety, and in some cases form, using Griffioen's (1995) key. A total of 35 Chrysanthemoides samples were used (Table 2). Additional sequence data for several species of Osteospermum (sections Homocarpa T. Norl. and Coriacea T. Norl.) were selected in order to test generic monophyly. Only species from these sections were considered, as other studies have indicated a close relationship between Chrysanthemoides and these sections of Osteospermum (Wood and Nordenstam, 2003;Nordenstam and Källersjö, 2009). It should be noted that O. subulatum DC. (of sect. Trialata in Norlindh, 1943) and O. triquetrum (unassigned to section in Norlindh, 1943) were transferred to sect. Homocarpa by Wood and Nordenstam (2003).
The more distantly related Tripteris microcarpa Harv. was used as outgroup.
Leaf samples were collected into silica gel according to the method of Chase and Hills (1991), and the DNA was extracted using a modified CTAB DNA extraction protocol (Doyle and Doyle, 1987). The ITS region was amplified using "ITS-5" (White et al., 1990) and a modified "ITS-4" primer ("ITS- Unless otherwise indicated, all vouchers are housed in the Selmar Schonland herbarium (GRA). An ⁎ indicates those samples which are incongruent between the ITS and cpDNA phyologenies (i.e. they swap positions between Clades 1 and 2). Numbers in the left column correspond to localities numbered on the map provided in Chrys-4"; 5′-TCCTCCGCTTATGGATATGC-3′). The psbA-trnH spacer was amplified using the primers "psbA" and "trnH" (Sang et al., 1997), and the trnL-F region amplified using the primers "c" and "f" (Taberlet et al., 1991). PCR amplifications were carried out using either a Thermo-Hybaid PCRSprint Temperature Cycling System or a Corbett Research PC-960G Microplate Gradient Thermal Cycler, with 35-40 cycles of amplification. Successful PCR amplification was confirmed by electrophoresing the PCR products on a 1% agarose gel. PCR products were cleaned using either the QIAGEN QIAquick PCR purification kit or the PROMEGA Wizard SV Gel and PCR purification kit and resuspended in 30 µl of dH 2 O before being sequenced using the ABI prism BigDye Terminator v3.0 or v3.1 Ready Reaction Cycle sequencing kit (Applied Biosystems) according to the manufacturer's instructions.
Sequence data was checked and edited using SEQUENCHER (Version 3.1.1; Gene Code Corporation). Assembled sequences were imported into MACLADE (Version 4.06; Maddison and Maddison, 2000) and aligned manually.

Phylogenetic analyses
As the psbA-trnH and trnL-F are both found on the chloroplast genome which is inherited as a single unit, these data sets were combined to form what we term the cpDNA data set. Prior to analysis, the incomplete 5′ and 3′ ends of the psbA-trnH and trnL-F regions were excluded. The ITS nrDNA data set was analysed separately. Data sets were analysed by means of Bayesian inference (BI) and maximum parsimony (MP). As Bayesian analysis is based on explicit models of DNA evolution, the program MrModelTest (Nylander, 2004) was used to select the model of DNA substitution that best fit the data. The Bayesian analysis was run using MrBayes v3.1.2 (Huelsenbeck and Ronquist, 2001) as follows: four Markov chains, three heated and one cold, were run simultaneously for 2,000,000 generations and trees were saved every 100 generations. The starting tree was random, the branch lengths were saved and the first 4000 trees were discarded as burn-in following a visual inspection of a plot of the likelihood values to ensure stationarity had been reached well within this limit. The remaining trees were combined and used to generate a 50% majority rule consensus tree and to determine the PP for each node.
Parsimony analyses were conducted on the two data sets as follows: Using PAUP version 4.0 b10, one hundred random input analyses were conducted using the parsimony-informative characters, keeping one tree (TKEEP = 1) from each of the analyses. A heuristic search was then conducted on all the shortest trees in memory (with MAXTREES set 20,000) using the TBR branch swapping algorithm. A strict consensus tree was then calculated from the set of equally parsimonious trees. A FULL HEURISTIC Bootstrap analysis was conducted on each data set using 1000 replicates, with MAXTREES set to 2000.

Results
MrModelTest identified the best DNA substitution model for the ITS data as the HKY + G (Hasegawa et al., 1985) model with variable sites assumed to follow a discrete gamma distribution. For the cpDNA data, the best DNA substitution model was identified as GTR+G. The Bayesian Inference consensus topologies for the ITS and cpDNA data sets are presented in Fig. 1, with the parsimony bootstrap values also indicated. The parsimony analyses produce poorly resolved consensus trees, and the nodes that are retained in common with the BI trees are indicated on Fig. 1. Table 3 lists the details of the parsimony analysis of each data set. Support for the various nodes in both trees is generally lacking, if one is to accept that any Bayesian posterior probability (PP) value b 0.95 and bootstrap percentages b 70% is weak (Alfaro and Holder, 2006).
The ITS region is twice as variable and contains approximately three times as many parsimony-informative sites as the cpDNA regions (Table 3). Despite this lower variability, the psbA-trnH region included three synapomorphic insertiondeletion events (indels) that, when mapped on the plastid tree, provide additional support for three of the main nodes retrieved ( Fig. 1; note the exception of a subsequent loss of one of these insertions in specimen SH100 -"C. monilifera subsp. pisifera var. pisifera form 2"within Clade 1). However, indel data were not coded and included in phylogenetic analyses.
It must be noted that all three samples of C. incana possessed multiple ITS paralogues of different lengths, which meant that not all the data from this region could be used, as the trace files became unreadable at the point where the paralogues diverged. This problem was particularly severe in the Wood 20 specimen, where 361 aligned sites were affected, and coded as unknown. In the remaining two C. incana samples, this problem affected less than 10 sites. Reasons for the presence of multiple paralogues in these samples include hybridization followed by incomplete lineage sorting, or the presence of pseudogenes. However, as the 5.8S regions of these sequences are highly conserved, it seems unlikely that the sequences used here are pseudogenes (Razafimandimbison et al., 2004). The obvious solution to this problem would be to clone the PCR product and sequence the different paralogues. While this has been done for a larger study based on ITS data only (Howis et al., in prep.), we used the data obtained from directly sequenced PCR product in this study. This finding is nonetheless interesting, and this species needs detailed molecular and morphological investigation.

Topological comparisons and incongruence
It is immediately apparent that the ITS and cpDNA topologies are not congruent, and in some instances this incongruence is well supported. The genus Chrysanthemoides is indicated by both datasets to be non-monophyletic, as in each tree there are various species of Osteospermum embedded within the Chrysanthemoides clade. Furthermore, the two additional lineages, representing the Drakensberg and East African taxa ("C. monilifera subsp. canescens" and "C. monilifera subsp. septentrionalis"), are placed in a more basal position, among species of Osteospermum in the ITS analysis. Both data sets resolve "C. incana subsp. incana" as a monophyletic species which is embedded within C. monilifera.
The two main clades of samples on the plastid phylogeny receive at best moderate support (pp =0.93 and BS = 84% for Clade 2). The ITS analysis shows higher levels of support (pp= 0.99, BS= 84% for Clade 1 and pp =0.92 for Clade 2, which has no BS support). Furthermore, each of these clades corresponds reasonably well with the Griffioen's taxonomy. Within Clade 1, the samples of "C. incana subsp. incana" are sister to the clade containing the remaining samples of "C. monilifera subsp. pisifera var. pisifera form 1" and "C. monilifera subsp. rotundata". Clade 2 comprises samples of "C. monilifera subsp. floribunda form 1", "C. monilifera subsp. monilifera", "C. monilifera subsp. pisifera var. pisifera form 2" and "C. monilifera subsp. pisifera var. angustifolia". In addition, the cpDNA places the two montane subspecies ("C. monilifera subsp. canescens" and "C. monilifera subsp. septentrionalis") in this clade as well. However, the ITS data places these taxa in a more basal position, among species of Osteospermum, but with no support.
These results indicate some (but not complete) agreement in terms of membership of each of the two clades. This incomplete agreement is further exacerbated by five samples of C. monilifera that swap clades between the two phylogenies. Three samples in Clade 1 of the cpDNA phylogeny are placed in Clade 2 of the ITS phylogeny (indicated by solid dots in Fig. 1), and two samples placed in Clade 2 of the cpDNA phylogeny are in Clade 1 of the ITS phylogeny (indicated by squares in Fig. 1).
Osteospermum ciliatum Berg. and O. aciphyllum are placed near the base of the tree in both phylogenies, while O. subulatum is in an unsupported clade of four Osteospermum taxa collectively sister to Clade 1 in the cpDNA phylogeny, but placed within Clade 1 of the ITS topology, with moderate to good support (Fig. 1). O. junceum and O. asperulum are sister taxa in the ITS topology, and are in turn sister to the "C. monilifera subsp. canescens" clade, a relationship that lacks support. However, in the cpDNA topology, O. junceum is sister to part of Clade 1 that comprises samples of C. monilifera (with moderate) and O. asperulum is sister to the rest of Clade 2 (with moderate support at best).
O. pyrifolium T. Norl. and O. triquetrum are members of Clade 2 of the cpDNA phylogeny, and placed in a well supported clade (pp = 0.97) with "C. monilifera subsp. canescens" samples. However, in the ITS phylogeny they placed as members of separate clades (O. pyrifolium is part of Clade 1 and O. triquetrum is sister to Clade 2). Wiens (1998) argues that conflicting data sets can be combined, and that the accuracy of the recovered topology as being a reflection of the true phylogeny is enhanced. We thus combined the data in an attempt to enhance the phylogenetic signal in the data. Parsimony analysis of the combined data resulted in a poorly resolved tree (nodes retained in the strict consensus tree are indicated in Fig. 3), a result typical of instances where hybridization has occurred and reduced resolution in parsimony analyses (McDade, 1992). However, the results of the BI analysis are encouraging, in that most samples of C. monilifera formed monophyletic lineages corresponding to their taxonomic identity (Fig. 3), although BI posterior probability support for most nodes was still lacking.

Support for infraspecific taxa within C. monilifera
While our sample size here is limited, the ITS sequence data does provide some evidence to support the recognition of the Fig. 1. Bayesian consensus trees from analysis of cpDNA (right) and nrDNA (left) data sets of 31 Chrysanthemoides samples and eight Osteospermum species. Numbers above the branches indicate posterior probabilities, and numbers below the branches are parsimony bootstrap values. Those nodes that are retained in the parsimony consensus trees are indicated as thick branches. The vertical bars indicate the two main clades of Chrysanthemoides samples discussed in the text. Osteospermum species are in bold, and the samples outlined in solid or dashed lines are the Afromontane subspecies of C. monilifera. The solid circles and squares indicate samples that are incongruent within the two main clades of Chrysanthemoides. The grey vertical bar at the base of Clade 1 in the cpDNA tree represents a 10 base pair insertion in the psbA-trnH data (with a subsequent loss in one specimen, indicated by the grey circle), the open bar at the base of Clade two in the cpDNA tree represent a 4-base pair deletion in the psbA-trnH data, and the thin black bar at the node above this represents a 6-base pair deletion in the psbA-trnH data. Key to taxon abbreviations: C. m = C. monilifera; C. m. p. = C. monilifera subsp. pisifera; C. i. = C. incana. multitude of infraspecific taxa recognized by Griffioen (1995), although few nodes receive good support. It is unfortunate that the plastid data is too conservative to provide independent support for these lineages. It should also be borne in mind that gene trees do not equate to species trees (Doyle, 1992), and that infraspecific relationships may be reticulate rather than hierarchical in nature. If so, then our results represent efforts to place square pegs (reticulating taxonomic entities) into round holes (a hierarchical representation of relationships).
Our results (especially the combined analysis; Fig. 3) do however indicate some genetic support for the infraspecific entities recognized by Griffioen (1995) on the basis of morphology (which is a phenotypic representation of the nuclear genotype). This lends weight to their validity, and we thus feel that it is important to encourage their use by the end-users of taxonomies. We thus present the key features of these entities in Table 1. This result highlights the value of careful phenetic and ecological studies in variable species (Griffioen, 1995), which should accompany molecular phylogeographic studies, as all sources of data need to be considered in assessing species limits and the recognition of ESUs within species. The fact that other of Griffioen's (1995) taxonomic entities are not monophyletic should not be viewed negativelyit is entirely possible that paraphyletic taxa (such as "C. monilifera subsp. rotundata", "C. monilifera subsp. pisifera form 2" and "C. monilifera subsp. floribunda form 1") represent surviving ancestral lineages, out of which some of the monophyletic lineages have recently evolved, and/or lineage sorting has not yet reached completion in these taxa. Another possibility is that this is a side effect of relatively rare hybridization. This should, in particular, affect the plastid phylogenies, as they tend to give more categorical results, and so can be expected to be categorically wrong.

Correlation to distribution
When the geographic distribution of the Chrysanthemoides samples in each clade identified in Fig. 1 is mapped, it becomes apparent that Clades 1 and 2 are correlated to geographic distribution: Clade 1 comprises samples from the eastern portion of the distribution range, and Clade 2 the western (Fig. 2). The western clade (Clade 2) comprises specimens from the predominantly winter rainfall region, whereas Clade 1 covers the bi-seasonal (but predominantly summer) rainfall region. Interestingly, the five samples that exchange clade membership between the ITS and cpDNA phylogenies are all from the geographic region intermediate between the main distribution areas of the two main clades; the Tsitsikamma -Nature's Valley -Oudtshoorn area, suggesting that hybridization between the two main clades in this region cannot be ruled out. It is thus possible that gene flow between these lineages has taken place, and that the different positions of these five samples in the ITS phylogeny reflects the fixation of one particular set of parental paralogues via concerted evolutionmost likely those from the paternal source which are incongruent with the maternal cpDNA topology.
In the analysis of the combined data, three of the five samples that swap clade membership (those indicated in Fig. 1 by circles) are placed within a clade that comprises most samples from Clade 1 with good support (pp= 0.95), reflecting the results of cpDNA analysis. The other two samples are placed basal to a clade that comprises [Clade 1; O. subulatum], a position that approximates the ITS result. The geographic patterns based on the analysis of the combined data are thus even more striking, with the enlarged Clade 1 (which includes O. pyrifolium and O. subulatum and having a pp = 0.98) now assuming a distinct Eastern distribution, while the remaining samples form a Western lineage (Figs. 2 and 3).

Relationship of Chrysanthemoides to Osteospermum
On the basis of both ITS and cpDNA gene trees as well as the combined analysis, both Osteospermum and Chrysanthemoides are paraphyletic, bearing in mind that many placements are not well supported. If these gene trees are taken to be straightforward reflections of evolutionary history, the simplest nomenclatural and taxonomic solution to this problem is to subsume all Chrysanthemoides taxa as well as Osteospermum sect. Coriacea (O. junceum) into Osteospermum sect. Homocarpa, which would then be monophyletic. This would mean that the defining morphological characteristic of Chrysanthemoides (drupaceous fruit) would then have to be interpreted as having originated several times, followed by losses/reversals in those species of Osteospermum shown to be embedded within the lineages of Chrysanthemoides. The number of gains and losses depends on the phylogeny used: cpDNA or nrDNA. However, it would be most profitable to re-examine the fruit morphology and its ontogeny for all species in Osteospermum section Homocarpa, as there is recent evidence (Wood and Nordenstam, 2003) that fruits with a fleshy exocarp are not restricted to Chrysanthemoides as claimed by Norlindh (1943) and that exocarp features (including colour) may have evolved independently and repeatedly as adaptations to different dispersal strategies (bird vs. ant dispersal; cf. Wood and Nordenstam, 2003).

The role of hybridization in Osteospermum and Chrysanthemoides
Irrespective of proposed taxonomic and nomenclatural changes required to preserve generic monophyly, the issue of conflicting species-level relationships within this assemblage remains. This could be caused by several processes. Generally, such conflict is attributed to hybridization and/or incomplete lineage sorting as well as the questionable utility of ITS as a suitable nuclear marker (see for example Vriesendorp and Bakker, 2005;Mort et al., 2008). Given the numerous instances of incongruence in this group, we would have to invoke multiple instances of hybridization (and possibly subsequent introgression), such that the general "Chrysanthemoides" form or morphology is retained throughout in disparate lineages, such as "C. monilifera subsp. canescens" and "C. monilifera subsp. rotundata". Certainly the intermediate geographic distribution of samples that swap clade membership may be considered as evidence for past or ongoing hybridization. The report of a natural hybrid between Osteospermum potbergense A. R. Wood & B. Nord. and Chrysanthemoides monilifera is noteworthy in this Map showing sample sites for South African specimens of Chrysanthemoides. Key to shapes: squares indicate localities of C. monilifera samples in Clade 1; circles indicate localities of C. monilifera samples in Clade 2; stars indicate localities of C. monilifera samples which swap clade membership between cpDNA and ITS data sets; triangles indicate localities of C. monilifera subsp. canescens; diamonds indicate localities of samples of C. incana. The grey outline indicates samples that fall into the Eastern clade in the analysis of the combined data set. Numbers within symbols indicate samples as listed in Table 1. context (Wood and Nordenstam, 2003). Alternatively, the entire Osteospermum section Homocarpa is of very recent origin, and the results here represent incomplete lineage sorting, a scenario that cannot be examined fully until a more complete phylogeny is obtained and subjected to some form of dating analysis.

Conclusion
Our results indicate the existence of substantial intraspecific variation within C. monilifera, corresponding in part to the various morphotypes recognized by Griffioen (1995) at different Fig. 3. Bayesian consensus tree from analysis of the combined (cpDNA and nrDNA) data set of 31 Chrysanthemoides samples and eight Osteospermum species. Numbers above the branches indicate posterior probabilities, and numbers below the branches are parsimony bootstrap values. Those nodes that are retained in the parsimony consensus tree are indicated as thick branches. The vertical bars indicate the two main clades of Chrysanthemoides samples discussed in the text. Osteospermum species are in bold, and the samples outlined in solid or dashed lines are the Afromontane subspecies of C. monilifera. The solid circles and squares indicate samples that are incongruent between the cpDNA and ITS data sets. Key to taxon abbreviations: C. m = C. monilifera; C. m. p. = C. monilifera subsp. pisifera; C. i. = C. incana. taxonomic ranks. Unfortunately, the taxonomic recognition and disentangling of these entities, along with the relationships of these to each other and other Osteospermum species is bedeviled by incongruence between nuclear and plastid markers, suggesting a history of hybridization events and reticulation or incomplete lineage sorting further confused by the distinct possibility of multiple paralogues of the ITS region. Despite the slight improvement in the phylogeny obtained when the data are combined, we refrain from nomenclatural validation of Griffoen's unpublished infraspecific taxa.

Significance to users of taxonomies
Our results are of considerable significance for biocontrol scientists working to control Chrysanthemoides in regions where it is invasive. As the current taxonomy does not adequately address the genetic diversity within species of Chrysanthemoides, it is essential for biocontrol scientists to ascertain which genetic lineage and geographic area the invasive plants are from e.g. Zachariades et al. (2004) and Paterson et al. (2009). Once this is known, it may be possible to obtain more effective biocontrol organisms from natural populations of that specific genotype, as for example shown for the control of the invasive fern Lygodium microphyllum by phytophagous mites (Goolsby et al., 2006).
Our findings are also relevant to the horticultural and landscaping industry, as C. monilifera is used for a range of horticultural practices (including landscaping and restoration, noted above). As we show that genetic diversity is correlated with geographic origin, it is clearly important to ensure that only locally adapted lineages are used for these activities, otherwise it is possible that foreign genotypes will be imported and may interbreed with local genotypes. This can have disastrous genetic consequences such as outbreeding depression and genetic swamping (Hufford and Mazer, 2003).