HOME HELP FEEDBACK SUBSCRIPTIONS ARCHIVE SEARCH TABLE OF CONTENTS		QUICK SEARCH:   [advanced] Author: Keyword(s): Year:  Vol:  Page:

This Article Abstract Full Text (PDF) Submit a response Alert me when this article is cited Alert me when eLetters are posted Alert me if a correction is posted Services Email this article to a friend Similar articles in this journal Similar articles in ISI Web of Science Alert me to new issues of the journal Download to citation manager Citing Articles Citing Articles via HighWire Citing Articles via ISI Web of Science (4) Citing Articles via Google Scholar Google Scholar Articles by Calviño, C. I. Articles by Downie, S. R. Search for Related Content PubMed Articles by Calviño, C. I. Articles by Downie, S. R. Agricola Articles by Calviño, C. I. Articles by Downie, S. R.

A molecular phylogenetic study of southern African Apiaceae¹

Department of Plant Biology, University of Illinois at Urbana-Champaign, Urbana, Illinois 61801 USA; ³Instituto de Botánica Darwinion, Buenos Aires, Argentina; and ⁴University of Johannesburg, Auckland Park, South Africa

Received for publication March 19, 2006. Accepted for publication October 5, 2006.

ABSTRACT

It has been suggested that southern Africa is the origin of the predominantly herbaceous Apiaceae subfamily Apioideae and that the woody habit is plesiomorphic. We expand previous molecular phylogenetic analyses of the family by considering all but three of the approximately 38 genera native to southern Africa, including all genera whose members, save one, have a woody habit. Representatives of five other genera are included because they may be closely related to these southern African taxa. Chloroplast DNA rps16 intron and/or nuclear rDNA ITS sequences for 154 accessions are analyzed using maximum parsimony, Bayesian, and maximum likelihood methods. Within Apioideae, two major clades hitherto unrecognized in the subfamily are inferred. The monogeneric Lichtensteinia clade is sister group to all other members of the subfamily, whereas the Annesorhiza clade (Annesorhiza, Chamarea, and Itasina) plus Molopospermum (and Astydamia in the ITS trees) are the successive sister group to all Apioideae except Lichtensteinia. Tribe Heteromorpheae is expanded to include Pseudocarum, "Oreofraga" ined., and five genera endemic to Madagascar. The southern African origin of subfamily Apioideae is corroborated (with subsequent migration northward into Eurasia along two dispersal routes), and the positions of the herbaceous Lichtensteinia and Annesorhiza clades within the subfamily suggest, surprisingly, that its ancestor was herbaceous, not woody.

Key Words: Apiaceae • Apioideae • cpDNA rps16 intron • phylogeny • rDNA ITS • Saniculoideae • southern Africa • woodiness

Southern Africa stands out by its great biological distinctiveness. The area customarily treated for floristic purposes as southern Africa includes Botswana, Lesotho, Namibia, South Africa, and Swaziland. The African subcontinent is remarkable for both its high plant richness and endemism at all taxonomic levels (Goldblatt, 1978 ; Goldblatt and Manning, 2002 ). It is also the origin, as well as the center of radiation, of numerous, diverse lineages of flowering plants. Approximately 53% of the total flora of the Cape region of South Africa has a shrubby habit, an overwhelming proportion even when compared with other Mediterranean floras such as those of California or Chile (Goldblatt and Manning, 2002 ). While the prevalence of woody species on oceanic islands has long attracted the attention of evolutionary biologists, less is known about the origin of the herbaceous vs. woody habit in continental floras, especially in the floristically unique southern Africa. It is often assumed that herbaceous plants within a family are derived from woody ancestors; such assumptions, however, have generally relied on implicit hypotheses of character-state polarity. It is of evolutionary interest to ascertain if the prevalence of woodiness in the southern African flora is the result of phylogenetic heritage or adaptation.

The immense phylogenetic importance of those plants native to southern Africa is exemplified by the family Apiaceae. Burtt (1991) , Van Wyk (2001) , and Van Wyk and Tilney (2004) emphasized the significance of these plants in terms of their unique morphological diversity and the difficulties in finding obvious, close relatives for many of them. Some of these genera have been postulated to be links between subfamilies Apioideae and Saniculoideae (Burtt, 1991 ); others may have evolved independently from these subfamilies (Van Wyk, 2001 ). The many unusual and defining features of these African plants include an arborescent or shrubby habit, deciduous leaves with dentate-aristate margins, and heteromorphic mericarps (Liu et al., 2003 ; Liu, 2004 ; Van Wyk and Tilney, 2004 ). Early molecular phylogenetic investigations revealed that two of these genera, Heteromorpha Cham. & Schltdl. and Anginon Raf., arborescent and shrubby plants endemic to sub-Saharan and southern Africa, either constitute a clade sister group to all other taxa of subfamily Apioideae or comprise successively basal-branching lineages within the subfamily (Downie et al., 1996 , 1998 ; Plunkett et al., 1996a , b ). The phylogenetic placements of these taxa suggested that southern Africa is the likely origin of the largely north temperate subfamily Apioideae and that the woody habit is plesiomorphic within the subfamily. Following these studies, Downie and Katz-Downie (1999) showed that the African endemic genera Anginon, Dracosciadium Hilliard & B. L. Burtt, Glia Sond., Heteromorpha, and Polemannia Eckl. & Zeyh. unite as a well-supported monophyletic group (named the Heteromorpha clade and, subsequently, tribe Heteromorpheae; Downie et al., 1998 , 2000b ). They further revealed that the woody apioid African endemic genera, Polemanniopsis B. L. Burtt and Steganotaenia Hochst., comprise a clade sister group to subfamily Saniculoideae, suggesting that this subfamily may have also evolved from woody ancestors. In the neighbor-joining and maximum likelihood trees inferred from phylogenetic analyses of chloroplast DNA (cpDNA) rps16 intron sequences, tribe Heteromorpheae is sister group to all other Apioideae (Downie and Katz-Downie, 1999 ). The strict consensus tree inferred from maximum parsimony analysis of these data, however, suggests that either tribe Heteromorpheae or the species Annesorhiza altiscapa Schltr. ex H. Wolff (or both) may be a sister group to a clade comprising the other members of subfamily Apioideae. Annesorhiza Cham. & Schltdl. is a genus of perennial herbs endemic to southern Africa, thus its position as a possible sister group to all other Apioideae is intriguing for it suggests a herbaceous ancestry for the subfamily, contrary to what has been generally assumed.

The importance of southern African Apiaceae in the early evolutionary history of the family is now well established, but knowledge of these umbellifers is limited by inadequate sampling, especially of the herbaceous African endemics. The relationships of the herbaceous taxa to the woody African genera and to the more northern herbaceous members of the family are not clear, and explicit hypotheses that include a broad representation of these woody and herbaceous African genera are lacking. Therefore, increasing the sampling of both herbaceous and woody southern African umbellifers (and including several non-African genera considered as belonging to the earliest diverging lineages of Apioideae) is critical in illuminating the direction of evolution of their unusual characters (many of which are hypothesized to be plesiomorphic within the family; Van Wyk and Tilney, 2004 ) and the historical biogeography of subfamily Apioideae.

The major objectives of this study are to (1) ascertain the phylogenetic placements of native southern African Apiaceae based on cpDNA rps16 intron and/or nuclear ribosomal DNA (rDNA) internal transcribed spacer (ITS) sequences, with emphasis on those genera whose members have a woody habit; (2) corroborate a southern African origin of subfamily Apioideae, as suggested by previous molecular systematic investigations, and reconstruct its subsequent biogeographic history; and (3) test the prevailing hypothesis that the woody habit is plesiomorphic in the subfamily. Progress on reclassifying the subfamily Apioideae, as well as having a better understanding of its origin and early diversification, can only be achieved upon consideration of these extraordinary African plants.

MATERIALS AND METHODS

Taxa and outgroup selection
A total of 154 accessions was examined for cpDNA rps16 intron and/or nuclear rDNA ITS sequence variation. In the phylogenetic analysis of rps16 intron sequences, 130 accessions were considered, which included 60 genera and 116 species of Apiaceae and two genera (three species) of Araliaceae. DNA sequences from 54 of these accessions were obtained specifically for this study (Table 1); data for the remaining 76 accessions were obtained during a previous study of woody southern African taxa (Downie and Katz-Downie, 1999 ). In the ITS phylogenetic analysis, 84 accessions of Apiaceae (representing 32 genera and 66 species) were considered and of these, 58 are new (Table 1). Voucher data and GenBank accession numbers for the 26 accessions examined previously for ITS sequence variation are available elsewhere: Komarovia Korovin and Physospermum Cusson (Downie et al., 1998 ); Aulacospermum Ledeb., Eleutherospermum K. Koch, Erigenia Nutt., Hansenia Turcz., Parasilaus Leute, and Pleurospermum Hoffm. (Katz-Downie et al., 1999 ); Molopospermum W. D. J. Koch (Downie et al., 2000a ); "Oreofraga" ined., Physospermopsis H. Wolff, and Pleurospermum (Downie et al., 2000c ); Cyclorhiza M. L. Sheh & R. H. Shan, Notopterygium H. Boissieu, Tongoloa H. Wolff, and Trachydium Lindl. (Valiejo-Roman et al., 2002a ); Anginon and Bupleurum L. (Neves and Watson, 2004 ); and Haplosphaera Hand.-Mazz. and Sinolimprichtia H. Wolff (Valiejo-Roman et al., 2006 ). Sixty accessions were common to both cpDNA and ITS data sets.

View this table: [in this window] [in a new window]   Table 1. New accessions of Apiaceae from which cpDNA rps16 intron and/or nuclear rDNA ITS sequences were obtained, with corresponding DNA accession and GenBank reference numbers and voucher information. Herbarium acronyms are according to Holmgren et al. (1990) . References to previously published rps16 intron and ITS sequences are cited in the text. Accessions Chamarea gracillima 2597 and C. longipedicellata each had ITS sequence heterogeneity and were not included in the phylogenetic analysis, nor were they submitted to GenBank

Representatives of five other genera were also considered, as prior studies or unpublished phylogenetic analyses suggested their close relationships to tribes Pleurospermeae or Heteromorpheae. These taxa are Molopospermum peloponnesiacum (L.) W. D. J. Koch (Downie et al., 2000a ), "Oreofraga morrisiana" ined. (Downie et al., 2000c ), Chamaesium paradoxum H. Wolff (Spalik and Downie, 2006 ), Pseudocarum C. Norman spp. (Van Wyk et al., 1999 ; Van Wyk, 2001 ), and Astydamia latifolia (L. f.) Baillon (K. Spalik, University of Warsaw, unpublished data). Molopospermum is a monotypic genus, with a limited distribution in montane and subalpine zones of the Pyrenees, Massif Central, and southern Alps; "Oreofraga morrisiana" (M. F. Watson and E. L. Barclay, Royal Botanic Garden Edinburgh, unpublished data) is a yet to be described species from Socotra; Chamaesium H. Wolff is a genus of small, montane, perennial herbs distributed from the east Himalayas to southwest China; Pseudocarum is a suffrutescent climber native to tropical east Africa and Madagascar; and Astydamia DC. is endemic to the Canary Islands. We included additional representatives of tribe Pleurospermeae, as well as Erigenia and members of the Komarovia clade, because these taxa comprise lineages adjacent to tribes Heteromorpheae and Bupleureae in previous phylogenetic analyses (Downie et al., 2001 ; Valiejo-Roman et al., 2002a ).

All trees resulting from analyses of rps16 intron sequences were rooted with Aralia L. and Hydrocotyle L. of the family Araliaceae (Chandler and Plunkett, 2004 ). As additional outgroups, we included members of Apiaceae subfamilies Azorelloideae and Mackinlayoideae (Chandler and Plunkett, 2004 ; Plunkett et al., 2004 ), primarily to help place the genus Hermas, which has been traditionally treated as a hydrocotyloid. Rooting of the ITS trees was not as straightforward because of high sequence divergence across the range of taxa considered. Among the various coding and noncoding loci utilized most commonly for Apiaceae phylogenetic study, the ITS region is the most rapidly evolving (Downie et al., 2001 ). At deep-level analyses within the family, however, its high rate of nucleotide substitution and many small insertion–deletion events (indels) make the alignments particularly problematic, confounding hypotheses of positional homology (Downie and Katz-Downie, 1996 ; Downie et al., 1998 , 2000c , 2001 ). Additionally, at this level of comparison, optimization alignment of ITS sequence data with respect to gap and mismatch weighting can result in different phylogenetic estimates (Petersen et al., 2002 ; Hardway et al., 2004 ). Pairwise sequence divergence values of between 5% and 15% among the taxa compared usually indicate an appropriate rate of divergence that will often yield readily alignable sequences (Olmstead and Palmer, 1994 ). Sequences that are substantially more divergent, particularly those with numerous indels, are difficult to align. For those regions of questionable alignment, it is best to exclude them from the analysis (Swofford and Olsen, 1990 ). The results of the rps16 intron phylogenetic analyses were used to partition the taxa sequenced for the ITS region into three data sets of supposedly related taxa so that ambiguity of alignment is reduced or avoided within each. These three ITS data sets comprised (1) tribe Heteromorpheae, with one accession of Annesorhiza (Annesorhiza clade) selected as the outgroup; (2) tribe Pleurospermeae and the Komarovia clade and its allies, with Bupleurum (tribe Bupleureae) selected as the outgroup; and (3) the Annesorhiza clade and its allies, with Lichtensteinia (Lichtensteinia clade) identified as the outgroup. Alternative selections of outgroups, such as rooting the trees of the Heteromorpheae data set with Bupleurum or Lichtensteinia or rooting the trees of the Annesorhiza clade data set with Heteromorpha, did not appreciably change the ingroup tree topologies.

Experimental strategy
Leaf material for DNA extraction was obtained directly from the field, from plants cultivated from seed in the greenhouse, from accessioned plants from botanical gardens, or from herbarium specimens. Voucher information is given in Table 1 for all newly obtained sequences. Total genomic DNA was extracted using the modified hexadecyltrimethylammonium bromide (CTAB) protocol of Doyle and Doyle (1987) as detailed in Downie and Katz-Downie (1996, 1999) or using a DNeasy Plant Mini Kit (Qiagen, Valencia, California, USA).

The region encompassing the rps16 intron and a portion of its flanking 3' exon was PCR-amplified using primer pair 5'exon-N (CGTTTGAAACGATGTGGTAG) and rps16 3'exon (CCTGTAGGYTGNGCNCCYTT) or 5'exon-C (TTTGAAACGATGTGGTAGA) and 3'exon-CR (ACCCACGTTGCGAAGAT). All primers are written 5' to 3'. The first pair is that of Downie and Katz-Downie (1999) , but primer 5'exon-N is six bases upstream from the original primer-binding site. The other primer pair was designed from a consensus sequence of previously published rps16 Apiaceae sequences to minimize self-dimer, hairpin, and pair-dimer formation. Internal primers rps16-C (TAAGAAGCACCGAAGTAATGTC) and rps16-CR (AATGGCGTTTCCTTGTTC) were designed specifically to facilitate some amplifications. PCR amplification methods are the same as described previously (Downie and Katz-Downie, 1996 , 1999 ). For template purification, the QIAquick PCR Purification or the QIAquick Gel Extraction Kits (Qiagen) were used following the manufacturer's instructions. Sequencing reactions were carried out using the BigDye Terminator version 3.1 Cycle Sequencing Kit (Applied Biosystems, Foster City, California, USA). Each of these reactions consisted of 2–5 µl purified PCR product, 5.2 µl of 12.5% glycerol, 2 µl of 5x sequencing buffer, 2 µl of 10 µM sequencing primer, 1.8 µl of sterile water, and 1 µl of Ready Reaction Mix (containing the dye terminators and AmpliTaq DNA polymerase). After an initial denaturation step of 1 min at 95°C, the following 35 thermal cycles were performed: (1) 15 s at 95°C, (2) 5 s at 45°C, and (3) 4 min at 60°C. All sequencing was done using an ABI (Applied Biosystems) 3730XL high-throughput DNA capillary sequencer at the Genetic Engineering Facility of UIUC's Biotechnology Center. The strategies employed to obtain the ITS sequence data are the same as presented elsewhere (Downie and Katz-Downie, 1996 ; Downie et al., 2000a ). With the exception of two accessions of Chamarea that had ITS sequence heterogeneity, all rps16 intron and ITS sequences acquired in this study were deposited in GenBank (Table 1).

Simultaneous consideration of both DNA strands across all sequenced regions permitted unambiguous base determination in all taxa, with two exceptions. The single accession of Chamarea longipedicellata demonstrated evidence of ITS sequence additivity at multiple nucleotide sites, as inferred by overlapping peaks on electropherograms from both forward and reverse sequencing runs. To examine the extent of sequence homogenization among reiterated ITS copies, molecular cloning of this species was conducted. Purified Taq polymerase-amplified PCR products were ligated to pCR4-TOPO plasmid vectors and subsequently transformed into chemically competent Escherichia coli cells using the TOPO TA Cloning Kit (Invitrogen, Carlsbad, California, USA) following the manufacturer's protocol. Sixty colonies were picked and cultured overnight on selective plates. Five transformed colonies were removed, boiled in 100 µl of sterile water for 15 min, then centrifuged at 4000 rpm for 5 min. To verify the presence of inserts in these transformed colonies, we made a 20-µl PCR cocktail containing 5.0 µl of supernatant, 2.0 µl of 10x Taq polymerase reaction buffer, 2 µl of 1.25 mM dNTPs, 1.2 µl of 50 mM MgCl₂, 0.5 µl each of 20 µM promoter primers T7 and M13 Reverse, 1 unit of Taq polymerase, and approximately 8 µl of sterile water. After an initial denaturation step of 10 min at 95°C, the following PCR cycling program was used: (1) 45 s at 95°C, (2) 60 s at 57°C, and (3) 60 s at 72°C. A 15-min 72°C extension period followed completion of 35 thermal cycles. PCR products were examined on 1% agarose gels for the presence of inserts, and concentrations were estimated by visual comparison with bands containing known amounts of DNA. Three positive transformants were purified using the QIAprep Spin Miniprep Kit (Qiagen) and the manufacturer's instructions, then sequenced using primer T7. One accession of Chamarea gracillima (2597) also had ITS sequence heterogeneity at multiple sites, but was not cloned.

Sequence comparisons and phylogenetic analyses
The determination of boundary sequences for the six major structural domains of the cpDNA rps16 group II intron was based on similar boundary sequences inferred for tobacco, mustard, and other Apiaceae (Michel et al., 1989 ; Neuhaus et al., 1989 ; Downie and Katz-Downie, 1999 ). Both intron and ITS sequences were aligned initially using the default pairwise and multiple alignment parameters in the computer program CLUSTAL X (gap opening cost = 15.00, gap extension cost = 6.66, DNA transition weight = 0.50; Jeanmougin et al., 1998 ) and realigned manually as necessary. Gaps were positioned to minimize nucleotide mismatches. In the alignment of rps16 intron sequences, gaps of equal length in more than one sequence were coded as the same presence or absence character state if they could not be interpreted as different duplication or insertion events. Similarly located but different length indels were coded as multiple binary characters. In several regions, gap coding was particularly problematic because of poly-A's, -G's, or -T's or indirect duplications of adjacent elements in two or more taxa. These gaps were not scored and these ambiguous regions were excluded from subsequent phylogenetic analyses. In the ITS alignments, gaps were not scored as additional binary characters because very few were informative among the ingroup taxa. Uncorrected pairwise nucleotide distances were calculated by PAUP* version 4.0b10 (Swofford, 2002 ). Sequence characteristics were obtained for each of the six structural domains of the cpDNA rps16 intron, as well as from the entire intron plus the flanking 3'-exon portion. Similar data were obtained for the three ITS data matrices.

Sequence data from both loci were incomplete for several accessions, and these regions were scored as missing in the analyses. Data for three regions (of 23, 36, and 76 bp) within the Chamaesium paradoxum rps16 intron could not be obtained despite repeated efforts; moreover, the sequence data we did procure were generally of poor quality, attributable to low DNA concentration. For the two smallest of these regions, data were also unobtainable for Chamarea gracillima and Pseudocarum eminii. Several additional accessions were missing data (of 5–50 bp) from both ends of the matrix. The Choritaenia capensis rps16 intron had long stretches of sequence additivity, as inferred by overlapping peaks on the electropherograms. Re-extracting its DNA and repeating the PCR and DNA sequencing did not yield better results; thus this species was excluded from the analysis of intron sequences. In the ITS study, sequence data were unavailable in GenBank for the 5.8S rDNA region for all accessions of tribe Pleurospermeae, the Komarovia clade and its allies, and for Erigenia and "Oreofraga." This region had little to no variation in the other accessions examined; hence, these missing data did not affect the phylogenetic results. Data for the ITS-2 region could not be obtained for Glia prolifera 2511 and Pseudocarum laxiflorum 167 despite our repeated but unsuccessful attempts to PCR-amplify this region.

The rps16 intron and ITS data matrices were analyzed separately using maximum parsimony (MP) as implemented by PAUP*. For the MP analysis of rps16 intron data (with and without informative gaps scored as binary characters), 1000 heuristic searches were initiated using random addition starting trees with tree-bisection-reconnection (TBR) branch swapping and MulTrees selected, but saving no more than five trees from each search. These trees were subsequently used as starting trees for further TBR branch swapping. The maximum number of saved trees was set to 20 000 and these were permitted to swap to completion. The strict consensus of these 20 000 minimal length trees was then used as a topological constraint in another round of 1000 random addition replicate analyses but, in this case, only those trees that did not fit the constraint tree were saved. No additional trees were found at the length of the initial shortest trees, suggesting that the strict consensus tree adequately summarizes the available evidence, even though the exact number of trees at that length is not known. Bootstrap values (Felsenstein, 1985 ) were calculated from 500 000 replicate analyses using "fast" stepwise-addition of taxa; only those values compatible with the 50% majority-rule consensus tree were recorded. For each of the three ITS data sets, heuristic MP searches were replicated 1000 times with random stepwise-addition of taxa, TBR branch swapping, and saving multiple trees. Bootstrap values were calculated from 1000 replicate analyses using TBR branch swapping and simple stepwise-addition of taxa. To examine the extent of conflict between the rps16 intron and ITS data sets for a comparable set of taxa, the incongruence length difference (ILD) test of Farris et al. (1995) was implemented using the partition homogeneity test of PAUP*. This test was carried out with 1000 replicate analyses, using the heuristic search option with simple addition of taxa and TBR branch swapping. Although serious questions have been raised regarding the value of this test as a criterion for deciding whether data should be combined into a single phylogenetic analysis (e.g., Yoder et al., 2001 ; Barker and Lutzoni, 2002 ), it is still a widely used method to assess data heterogeneity and combinability. The number of additional steps required to force particular taxa into a monophyletic group was examined using the constraint option of PAUP*.

Bayesian inference of rps16 intron sequences was conducted using the program MrBayes version 3.0 (Huelsenbeck and Ronquist, 2001 ). The program was run in parallel on an IBM pSeries 690 system at the National Center for Supercomputing Applications at UIUC. Prior to analysis, the program Modeltest version 3.5 (Posada and Crandall, 1998 ) was used to select an evolutionary model of nucleotide substitution (among 56 possible models) that best fit these data, as selected by the Akaike Information Criterion estimator (Posada and Buckley, 2004 ). The settings appropriate for the best-fit TVM+G model were put into a MrBayes block in PAUP* (nst = 6; rates = gamma). The priors on state frequencies and rates and variation across sites (shape of the gamma distribution) were estimated automatically by the program. From a random starting tree, a Bayesian analysis was run for 7 million generations and the trees saved to a file every 100 generations (i.e., 70 000 trees were sampled). Sixteen simultaneous Markov chain Monte Carlo (MCMC) chains were run, and the temperature was adjusted to 0.05 in order to keep an appropriate heat range for the 16 chains. Branch lengths of the trees were saved. Variation in likelihood scores to determine apparent stationarity was examined graphically using the program Tracer version 1.2.1 (A. Rambaut and A. Drummond, University of Oxford, unpublished data). The states of the chain that were sampled before stationarity (i.e., the "burn in" of the chain) were discarded, and the posterior probability values for each bipartition of the phylogeny were determined from the remaining trees. MCMC convergence was also explored graphically using the cumulative option of the program AWTY online (Wilgenbusch et al., 2004 ), that displays posterior probabilities of splits at selected increments over a MCMC run.

The rps16 intron data set was also analyzed using the maximum likelihood method as implemented by PAUP* after the program Modeltest was used to select an appropriate model of nucleotide substitution. The results obtained were congruent to those inferred by the Bayesian analysis; hence they will not be discussed further.

Biogeographic analysis
To reconstruct the distribution of the ancestor of subfamily Apioideae, a dispersal–vicariance analysis was carried out with the program DIVA version 1.1 (Ronquist, 1996 ), using the optimize command and default option settings. We entered the following simplified, fully resolved phylogenetic tree based on the results of the rps16 intron analysis: (Azorelloideae, (Hermas clade, ((Steganotaenia & Polemanniopsis, Saniculoideae), (Lichtensteinia clade, ((Annesorhiza clade, Molopospermum), (Heteromorpheae, (Bupleureae, (Pleurospermeae, (Komarovia clade, (Oenantheae, apioid superclade and allies [the latter representing tribes Scandiceae, Aciphylleae and Smyrnieae])))))))))). With the exceptions of the Annesorhiza, Lichtensteinia and Hermas clades, which are newly recognized in this study, this tree is also congruent with a summary of relationships within subfamily Apioideae as revealed by phylogenetic analyses of seven molecular data sets (Downie et al., 2001 ). Because one of our primary goals was to corroborate a southern African origin of subfamily Apioideae, only three unit areas were defined: (a) southern Africa (i.e., Botswana, Lesotho, Namibia, South Africa, and Swaziland), (b) rest of the world, and (c) sub-Saharan Africa (excluding southern Africa). We coded each terminal taxon for its likely ancestral distribution and not for all of the regions in which its members presently occur, because if we had, information important for the optimization of ancestral states would be lost (Ronquist, 1996 ). In most cases, however, ascertaining the ancestral distribution of a terminal taxon was clear, because it was endemic to one of the defined unit areas. The likely ancestral distribution of the widespread tribe Bupleureae was determined based on a recent phylogenetic study by Neves and Watson (2004) , in which a Eurasian (specifically, a western Mediterranean) origin for the group was inferred. Members of tribe Heteromorpheae are distributed primarily in southern Africa, but some genera (i.e., Heteromorpha, Pseudocarum, and "Oreofraga") also extend northward to Ethiopia and Yemen (including Socotra) (Winter and Van Wyk, 1996 ; Van Wyk et al., 1999 ). Given these distributions and in an effort to determine how each ancestral area affected the reconstruction of the distribution of the ancestor of Apioideae, we ran three different analyses assuming that the ancestor of tribe Heteromorpheae was distributed only in southern Africa, only in sub-Saharan Africa, or in both places.

Evaluating the distribution of the woody habit
Within the independent context of the phylogenies inferred from MP analysis of rps16 intron sequences, the evolutionary pattern of woodiness was hypothesized within subfamily Apioideae. The character habit, scored as herbaceous or woody for each terminal and designated as unordered, was optimized onto all minimal length trees. The "State Changes & Stasis" chart of MacClade version 4.07 (Maddison and Maddison, 2005 ) was used to determine the minimal number of times the woody habit had evolved. These reconstructions of character evolution were displayed graphically using the option "Trace Character" in MacClade's tree window.

RESULTS

Comparisons and phylogenetic analyses of cpDNA rps16 intron sequences
Among the 130 sequences obtained, the rps16 intron varied in length from 758 (Diplolophium somaliense) to 908 bp (Annesorhiza altiscapa) and averaged 862 bp. Juxtaposed was an additional 17 bp of sequence data from the rps16 3'exon that were retained in the analysis because they contained phylogenetic information. Alignment of these intron and flanking exon sequences resulted in a matrix of 1332 positions, of which 278 were excluded from subsequent analyses because of alignment ambiguities. These ambiguous regions ranged from 1 to 43 bp in size. Of the remaining 1054 unambiguously aligned positions, 607 were not variable, 135 were variable but uninformative, and 312 were parsimony informative. Percentage G + C content for the intron ranged from 31.3 to 37.3%, averaging 34.3%. The g1 statistic for 10 000 random trees was –0.354. This value is significantly more skewed (i.e., more negative) than random data (g1 = –0.09 for 250 variable positions for more than 25 taxa; P < 0.01), indicating that these sequence data contain significant amounts of phylogenetic signal (Hillis and Huelsenbeck, 1992 ). A total of 125 unambiguous gaps, ranging between 1 and 126 bp, was required for proper alignment of these sequences; 61 gaps of 1–22 bp in size were parsimony informative. The frequency distribution of both informative and uninformative gaps in relation to their sizes is presented in Fig. 1. Relative to the outgroup Aralia chinensis L., these gaps represent a minimum of 84 insertion and 39 deletion events (two gaps could not be polarized because they were unique to Aralia). Pairwise sequence divergence estimates ranged from identity to 13.3% of nucleotides [the latter between Hydrocotyle rotundifolia Wall. and Torilis arvensis (Huds.) Link]. Each of the following groups of taxa yielded identical DNA sequences: two species of Bupleurum (B. chinense and B. americanum); three species of Eryngium L. (E. alternatum J. M. Coult. & Rose, E. mexiae Constance, and E. yuccifolium Michx.); four accessions of Anginon (A. difforme 2455, A. fruticosum 2583, 2610, and A. swellendamense 2463); two accessions of Heteromorpha arborescens var. arborescens (42, 2631); nine accessions of Anginon, Glia, and Polemannia (A. difforme 2582, A. intermedium, A. paniculatum, A. pumilum 2606, A. rugosum 513, 2607, A. verticillatum 2608, G. prolifera 1308, and P. grossularifolia); two accessions of Annesorhiza macrocarpa (2454, 2796); and the accessions Annesorhiza filicaulis and Chamarea snijmaniae.

View larger version (13K): [in this window] [in a new window]   Fig. 1. Frequency of unambiguous gaps in relation to gap size inferred in the alignment of 130 cpDNA rps16 intron and partial flanking 3'-exon sequences from Apiaceae and Araliaceae. Each bar represents the total number of parsimony informative (solid) and/or uninformative (shaded) gaps, with no overlap between them when they co-occur at the same size interval. Sixty-one gaps, of 1–22 bp in size, were parsimony informative; 64 gaps, of 1–126 bp in size, were uninformative

View larger version (42K): [in this window] [in a new window]   Fig. 2. Strict consensus of 20 000 minimal length 1148-step trees derived from equally weighted maximum parsimony analysis of 130 cpDNA rps16 intron and partial flanking 3'-exon sequences plus 61 binary-scored alignment gaps (CIs = 0.6028 and 0.5366, with and without uninformative characters, respectively; RI = 0.8820). Numbers at nodes are bootstrap estimates for 500 000 replicate analyses using "fast" stepwise-addition of taxa; values <50% are not indicated. The names of the major clades are based on previous studies or are newly recognized in this study

View larger version (43K): [in this window] [in a new window]   Fig. 3. Fifty-percent majority-rule consensus of 50 320 trees derived from Bayesian analysis of 130 cpDNA rps16 intron and partial flanking 3'-exon sequences. Numbers at nodes are posterior probability values. The names of the major clades are based on previous studies or are newly recognized in this study

The phylogenies inferred from rps16 intron sequence data help to clarify the phylogenetic placements of several taxa. Pseudocarum spp. allies with Dracosciadium, which is included along with Anginon, Glia, Heteromorpha, and Polemannia in tribe Heteromorpheae. Molopospermum is sister group to the Annesorhiza clade, but this relationship is not very strongly supported (59% bootstrap, 56% posterior probability). The five shrubby and endemic South African species of Peucedanum are closely related, with 0.1–1% pairwise sequence divergence among them, and in the Bayesian tree they form a clade with the South African genus Dasispermum. Chamaesciadium acaule, from the Caucasus, allies strongly with Aegokeras caespitosa from Turkey (95% bootstrap, 100% posterior probability) and is a new addition to tribe Careae. Deverra occurs in the Apium clade of the apioid superclade. The genera Steganotaenia and Polemanniopsis, treated traditionally in subfamily Apioideae, unite as a well-supported clade that is sister group to subfamily Saniculoideae. This sister group relationship, however, is supported weakly, with 58% and 87% bootstrap and posterior probability values, respectively. Constraining Steganotaenia, Polemanniopsis, and all accessions of subfamily Apioideae to monophyly in a subsequent MP search resulted in trees just one step longer than those without the constraint. In these suboptimal trees, Steganotaenia/Polemanniopsis is sister group to the clade comprised of Heracleum through Lichtensteinia.

The phylogenetic position of Chamaesium paradoxum remains unclear. Sequence data were only available for about 80% of its intron and much of it was of poor quality. The results of an initial MP analysis placed C. paradoxum on one branch of a four-branched polytomy in a strict consensus tree, along with tribe Heteromorpheae, the Annesorhiza clade plus Molopospermum, and the large, distal clade comprised of Heracleum through Bupleurum. In the majority-rule consensus tree, C. paradoxum is sister group to the Heracleum through Bupleurum clade. While these results indicate that Chamaesium paradoxum has allies among the basal apioids, its phylogenetic position is far from certain. Moreover, the poor quality of these data precluded their consideration in the final analysis of rps16 intron sequences.

Nuclear rDNA ITS sequence comparisons and phylogenetic analyses
Chamarea longipedicellata and one accession of C. gracillima (2597) demonstrated evidence of ITS sequence additivity at multiple nucleotide sites. Molecular cloning of the single accession of Chamarea longipedicellata revealed intra-individual ITS polymorphisms. Three sequence types were identified from the ITS-1 region, differing from each other by 7–18 nucleotide substitutions. Gene trees inferred by combining these three sequences with data obtained from direct sequencing of PCR products from other members of the Annesorhiza clade revealed the two most similar clones of C. longipedicellata comprising a clade with C. snijmaniae, a result in accordance with that inferred by phylogenetic analysis of rps16 intron sequences. The third clone allied with the two accessions of Annesorhiza altiscapa, differing from them by four nucleotide substitutions. These intra-individual paralogous ITS sequences plus the sequences obtained from direct sequencing of C. longipedicellata and C. gracillima (2597) were not included in subsequent analyses because they would mislead phylogenetic inferences.

Details of the alignments of the three ITS data sets of putatively related taxa are presented in Table 3. For each matrix, less than 11% of all positions were excluded because of alignment ambiguities. The first matrix comprised members previously attributable to tribe Heteromorpheae and, as a result of the rps16 intron analyses, two species of Pseudocarum. We also included the genus "Oreofraga" based on sequence similarity. When compared with any other member of this group, "Oreofraga" has a sequence divergence value of between 7.4% and 12.9% and is readily aligned. The second matrix included Erigenia bulbosa and those accessions of tribe Pleurospermeae and the Komarovia clade currently available in GenBank. We also included several Sino-Himalayan and other Asian species of Apioideae having a close affinity to taxa of the Komarovia clade (Valiejo-Roman et al., 2002a ). Diplolophium somaliense was included in this group, because it fell between the aforementioned clades in the rps16 intron trees, and so was Chamaesium paradoxum, because a possible affinity to tribes Bupleureae and Heteromorpheae was established in the rps16 intron analysis. Pairwise divergence estimates between C. paradoxum and the outgroup Bupleurum ranged from 5.8 to 7.5%; when C. paradoxum was compared to any other member of this matrix, divergence values ranged from 20.8 to 27.2%. Sequence divergence estimates between any ingroup member (other than C. paradoxum) and Bupleurum were high, with values ranging from 27.2 to 34.9%. A close relationship between C. paradoxum and Annesorhiza was also established in the intron analysis, but aligning C. paradoxum with any member of the Annesorhiza clade proved difficult, resulting in several long, ambiguously aligned regions. We acknowledge that the phylogenetic placement of C. paradoxum is unclear and that its inclusion in the second matrix is simply based on the similarity of its ITS sequence with those of Bupleurum. The third matrix included all representatives of the Annesorhiza clade plus Molopospermum, Astydamia, and Choritaenia. Molopospermum is a weakly supported sister taxon to the Annesorhiza clade in the rps16 intron trees, and the Astydamia ITS sequence was similar to those of Annesorhiza, Chamarea, and Molopospermum, with up to 11.4% maximum nucleotide divergence in pairwise comparisons. Across all ITS sequences, Choritaenia was most similar to others of this group; here, pairwise comparisons ranged from 14.3–18.6% (the latter between Choritaenia and Molopospermum).

View larger version (71K): [in this window] [in a new window]   Fig. 4. Strict consensus trees derived from MP analyses of nuclear rDNA ITS sequences of partitioned taxa: Tree A, tribe Heteromorpheae (no. of MP trees = 4; length of MP trees = 203 steps; CI without uninformative characters = 0.7438; RI = 0.9034); tree B, tribes Pleurospermeae and Erigenieae and the Komarovia and Physospermopsis clades (no. of MP trees = 9; length of MP trees = 661 steps; CI without uninformative characters = 0.5584; RI = 0.7372); tree C, Annesorhiza clade plus allies (no. of MP trees = 3; length of MP trees = 329 steps; CI without uninformative characters = 0.6752; RI = 0.7778). Bootstrap values are indicated on branches and were calculated from 1000 replicate analyses. The taxa for which ITS sequences were obtained in this investigation can be identified by their corresponding DNA accession numbers; those without numbers were obtained directly from GenBank. Complete taxon names, including the ranks of infraspecific taxa, are in Table 1

In the rps16 intron trees (Figs. 2 and 3), members of the Annesorhiza clade comprise a polytomy, with its largest resolved branches intermixing accessions of Annesorhiza and Chamarea. Annesorhiza filicaulis and Chamarea snijmaniae possess identical rps16 intron sequences and are sister species; this clade is sister group to C. longipedicellata. A similar scrambling of taxa occurs in the ITS strict consensus tree (Fig. 4C), but with Annesorhiza filicaulis now a weakly supported sister group to Itasina filifolia. Chamarea snijmaniae unites with three other Chamarea species in a strongly supported clade and not with A. filicaulis, as it does in the rps16 intron tree. Results of a partition homogeneity test on a set of 17 accessions common to both data sets (i.e., 12 members of the Annesorhiza clade, Molopospermum, and four accessions of Lichtensteinia) revealed that these matrices yield significantly different phylogenetic estimates (ILD probability value = 0.001). As a result of this significant discordance between data sets, these data were not combined for simultaneous analysis. Otherwise, reduced or erroneous resolution with respect to the true organismal phylogeny would result without further evaluation of the source of conflict.

Biogeographic analyses
The results of the three dispersal–vicariance analyses, assuming that the ancestor of tribe Heteromorpheae was distributed only in southern Africa, only in sub-Saharan Africa, or in both places (Fig. 5A–C, respectively), indicated that the likely ancestral distribution of subfamily Apioideae was in southern Africa. Two alternative biogeographic scenarios were obtained for its distribution, however. One scenario is that the ancestor of Apioideae was strictly distributed in southern Africa. The other scenario, suggesting a "rest of the world" distribution including southern Africa, was recovered as one of two alternatives when the ancestor of Heteromorpheae was inferred to have a strictly sub-Saharan African distribution (Fig. 5B). While we defined the unit area as "rest of the world," the distribution of taxa comprising tribes Bupleureae and Pleurospermeae and the Komarovia clade is largely Eurasian. In all analyses, the biogeographical reconstructions of the ancestor of Apioideae to the ancestor of the clade comprising the Apium superclade and allies through Bupleurum vary depending on the likely distribution of the ancestor of Heteromorpheae. Each reconstruction, however, postulates a migration north to Eurasia that would have required 2–3 dispersal events.

View larger version (28K): [in this window] [in a new window]   Fig. 5. Optimal reconstructions of the ancestral distributions of Apioideae using dispersal–vicariance analysis assuming that the ancestor of Heteromorpheae was distributed in (A) southern Africa only, (B) sub-Saharan Africa only, or (C) both places. At each node, the optimal distribution prior to vicariance is given; alternative and equally optimal distributions are separated with a space. The historical biogeography of the deepest splits within Apioideae was analyzed in terms of three main areas: southern Africa ("a" and solid bars); rest of the world ("b" and open bars); and sub-Saharan Africa ("c" and striped bars). An asterisk indicates the origin of subfamily Apioideae. Arrows represent dispersal events for one alternative reconstruction in each tree. Dispersal events are not indicated in Fig. 5B because of the complexity of the different scenarios

DISCUSSION

Subfamilies of Apiaceae and their relationships
Four subfamilies are currently recognized in Apiaceae: Apioideae, Saniculoideae, Azorelloideae (= Azorella clade of Downie et al., 2000b , 2001 ), and Mackinlayoideae (Plunkett et al., 2004 ). Apioideae and the clade of Saniculoideae plus Steganotaenia/Polemanniopsis comprise monophyletic sister groups, with subfamilies Azorelloideae (Azorella Lam., Bolax Comm. ex Juss., and Eremocharis Phil.) and Mackinlayoideae (Centella L. and Xanthosia Rudge) comprising successively more basal branching lineages. The traditionally recognized subfamily Hydrocotyloideae (sensu Drude, 1898 ) is polyphyletic, with some of its members occurring within Apiaceae subfamilies Azorelloideae and Mackinlayoideae, and others (such as Hydrocotyle) now included within Araliaceae (Chandler and Plunkett, 2004 ). The hydrocotyloid genus Klotzschia Cham., forming an isolated lineage in the rps16 intron trees, has been provisionally included in subfamily Azorelloideae (Downie et al., 2001 ; Plunkett et al., 2004 ). However, because of its distinctive fruits (Liu, 2004 ) and its consistent occurrence as a separate lineage in all studies where it is included, Klotzschia may very well comprise a monogeneric subfamily, pending examination of its type species K. brasiliensis Cham. & Schltdl. The four included accessions of Hermas, a genus endemic to the fynbos region of South Africa and heretofore not included in any molecular systematic study, form a strongly supported clade (100% bootstrap and posterior probability). In the Bayesian tree (Fig. 3) and the MP strict consensus tree without scored gap characters (not shown), the Hermas clade arises as a weakly supported sister group to Apioideae plus the clade of Saniculoideae and Steganotaenia/Polemanniopsis. When gaps are included (Fig. 2), the relationships among many of the basal branching lineages of Apiaceae are unresolved. Hermas is traditionally placed in the subfamily Hydrocotyloideae but presents several features, such as multiflowered congested umbels and a haploid chromosome number of seven that have been interpreted as characteristic of subfamily Saniculoideae (B. J. de Villiers et al., University of Johannesburg, unpublished data). Further studies of the polyphyletic subfamily Hydrocotyloideae are in order, especially to confirm the phylogenetic placements of Klotzschia and Hermas. The results obtained to date, however, are compelling in suggesting that upon further investigation, these genera may be treated as new subfamilies of Apiaceae or constitute members of an expanded subfamily Azorelloideae.

Steganotaenia and Polemanniopsis
Our results continue to support the earlier finding that Steganotaenia and the monotypic Polemanniopsis, arborescent and shrubby plants that were treated previously in subfamily Apioideae (e.g., Pimenov and Leonov, 1993 ), unite as a well-supported clade (85% bootstrap, 100% posterior probability) sister group to subfamily Saniculoideae (Downie and Katz-Downie, 1999 ). Each of the new accessions of Steganotaenia and Polemanniopsis examined yielded almost identical rps16 intron sequences to their previously sequenced congeners. Additional synapomorphies supporting the union between Steganotaenia and Polemanniopsis include the presence of large, distinct marginal fruit wings with enormous intrajugal cavities (Liu et al., 2003 ). Subfamily Saniculoideae (represented here by Astrantia L., Eryngium, Hacquetia Neck. ex DC., Petagnaea Caruel, and Sanicula L., and by Actinolema Fenzl, Alepidea, and Arctopus in other studies [Plunkett and Lowry, 2001 ; Valiejo-Roman et al., 2002b ; C. I. Calviño and S. R. Downie, unpublished data]) forms a strongly supported clade on the basis of molecular evidence (100% bootstrap and posterior probability). The monophyly of Saniculoideae is supported by numerous morphological features (Drude, 1898 ; Van Wyk, 2001 ; Liu et al., 2003 ). The placement of Steganotaenia and Polemanniopsis into an expanded Saniculoideae, as suggested by Downie and Katz-Downie (1999) and implemented by Liu et al. (2003) , cannot easily be reconciled on the basis of morphological data, nor does it receive strong support from the molecular analyses presented herein. While these taxa share the complete absence of commissural and vallecular vittae (oil ducts in the commissure and furrows, respectively), these vittae are also absent in Lichtensteinia and many hydrocotyloids. We are not aware of any morphological synapomorphy supporting the union of Steganotaenia/Polemanniopsis with Saniculoideae that does not also include Lichtensteinia and members of subfamilies Azorelloideae and Mackinlayoideae. The only evidence clearly justifying the union of Steganotaenia and Polemanniopsis with Saniculoideae is that of the rps16 intron; this relationship, however, is only supported weakly, with 58% bootstrap and 87% posterior probability values. Moreover, a close relationship to subfamily Apioideae cannot be completely ruled out, for Steganotaenia/Polemanniopsis and Apioideae are monophyletic in trees just one step longer than those maximally parsimonious. Further molecular systematic studies on Steganotaenia/Polemanniopsis are in progress, and the results should illuminate the proper phylogenetic position of these taxa (C. I. Calviño and S. R. Downie, unpublished data).

Lichtensteinia
The five accessions of Lichtensteinia comprise a strongly supported monophyletic group in the rps16 intron trees (94% bootstrap, 100% posterior probability). We recognize this group as the Lichtensteinia clade. The affinity of Lichtensteinia has long been problematic because it shares features common to both subfamilies Saniculoideae and Apioideae (Burtt, 1991 ; Van Wyk, 2001 ). Emphasizing its large compound umbels and compound leaves, the genus was placed in Apioideae (Drude, 1898 ; Wolff, 1910 ; Pimenov and Leonov, 1993 ). By focusing attention on its fruits instead, the genus was included in Saniculoideae (Koso-Poljansky, 1916 ; Liu et al., 2003 ). The fruits of Lichtensteinia are characterized by prominent intrajugal vittae (oil ducts in the ribs) and the absence of commissural and vallecular vittae. These same features also occur in subfamily Saniculoideae, the genera Steganotaenia and Polemanniopsis, and in the hydrocotyloid lineages (Wolff, 1913 ; Tseng, 1967 ; Liu et al., 2003 ). In contrast, almost all other genera in Apioideae have large commissural and vallecular vittae and only some have small intrajugal secretory ducts (Drude, 1898 ; Liu et al., 2003 ). The presence of large intrajugal secretary ducts and the lack of both commissural and vallecular vittae among basal members of Apiaceae represent plesiomorphic character states, and the inclusion of Lichtensteinia, Steganotaenia and Polemanniopsis within an expanded subfamily Saniculoideae, as suggested by Liu et al. (2003) , is based on these symplesiomorphies. Further studies are necessary to corroborate the results obtained here, in which Lichtensteinia is sister group to a clade comprising all other taxa of subfamily Apioideae. If this relationship is maintained, a new monotypic tribe of Apioideae will be warranted. Such confirmation of the phylogenetic position of Lichtensteinia is also important to enable hypotheses on character state evolution within the subfamily.

Annesorhiza clade
The Annesorhiza clade includes Annesorhiza, Chamarea, and Itasina and is well supported in analyses of both chloroplast and nuclear DNA sequences (98–100% bootstrap, 100% posterior probability). These genera are deciduous, perennial herbs endemic to southern Africa. Annesorhiza and Chamarea have been treated previously as sister taxa (Tilney and Van Wyk, 2001 ); they are highly similar morphologically, sharing features such as expanded and lignified vascular bundles in the fruit walls, hysteranthous leaves (i.e., leaves that are formed only after flowering and fruiting), heteromorphic fruits in some species, and fleshy pencil-like or tuberous roots that are replaced periodically (Van Wyk and Tilney, 1994 ; Tilney and Van Wyk, 2001 ; Liu, 2004 ). Annesorhiza and Chamarea differ in their overall size and the shape and relative length of their fruits, with Annesorhiza usually having larger leaves, taller scapes, and oblong-shaped fruits slightly longer than the rounded or flask-shaped fruits typical of Chamarea (Tilney and Van Wyk, 2001 ). In practice, however, these characters overlap, and the boundaries between these genera are unsatisfactory. Cladistic analysis of 29 anatomical and morphological characters from leaves, fruits, and roots revealed that the concept of Annesorhiza should be broadened to include all known species of Annesorhiza and Chamarea (Vessio, 2001 ). This finding is supported by molecular evidence in which these two genera do not comprise monophyletic sister groups. Instead, the three major clades resolved in the ITS strict consensus tree (Fig. 4C) correspond to the three sections of Annesorhiza recognized by Vessio (2001) , with the only difference being the inclusion of Chamarea sp. nov. 2594 (=Annesorhiza elsiae Vessio, Tilney & B-E. van Wyk) alongside members of sect. Annesorhiza in our study rather than with A. filicaulis in sect. Ternata in the treatment by Vessio (2001) .

Annesorhiza filicaulis was recently transferred into Peucedanum (as Peucedanum filicaule (Eckl. & Zeyh.) Van Wyk and Tilney) because its fruits have a wing configuration quite unlike that of any other species of Annesorhiza (Van Wyk and Tilney, 2001 ). In Annesorhiza, the four commissural ribs are somewhat larger than the other ribs, and this asymmetrical development is taken to an extreme in A. filicaulis in which they are expanded to form large wings, such as those typical of the genus Peucedanum (Tilney and Van Wyk, 2001 ; Van Wyk and Tilney, 2001 ). This transfer, however, was done so with some uncertainty, for its retention in Peucedanum would depend ultimately upon a revision of Peucedanum and related genera in southern Africa and elsewhere (Van Wyk and Tilney, 2001 ). The cpDNA-derived phylogenies show that A. filicaulis is sister group to Chamarea snijmaniae, with this clade in turn sister group to C. longipedicellata. These same trees also indicate that A. filicaulis is quite distantly related to the five South African accessions currently treated in Peucedanum, suggesting that the transfer of A. filicaulis into Peucedanum was premature. Vessio (2001) concurred that A. filicaulis should be maintained alongside Annesorhiza and Chamarea in an expanded Annesorhiza.

The relationship between Annesorhiza (12 species; Tilney and Van Wyk, 2001 ) and Chamarea (5 species; Van Wyk, 2000 ) is complex. Considering their similar morphologies and the significant discordance between the plastid- and nuclear-derived data sets, past hybridization/introgression or polyploidization seems possible, and further study of these taxa is warranted. Other than a chromosome count of N = 12 for A. macrocarpa (Burtt, 1991 ), the chromosome numbers are not known for any other member of the Annesorhiza clade, and they would be important to know to confirm polyploidy, if it exists. The presence of additive patterns of bands in the ITS electropherograms of two species of Chamarea may be associated with allopolyploidy and would be consistent with the retention of multiple rDNA loci from different parental ancestors at the time of speciation (Soltis et al., 1992 ). Although no interspecific hybridization has been reported for the largely sympatric Annesorhiza and Chamarea, hypotheses of relationships based on morphology show considerable incongruence in the data, with many species not very well separated (Tilney and Van Wyk, 2001 ). The high level of incongruence between cpDNA and nuclear rDNA suggests that hybridization and/or introgression may have been events in the early history of these plants, resulting in the similarities observed today among the species. While the extent of hybridization in the southern Africa flora is poorly known, numerous examples of naturally occurring hybrids have been reported (Goldblatt, 1978 ; Spies et al., 1987 ; Takatsu et al., 2001 ; Roodt and Spies, 2003 ). Given the diverse environments and recurring climatic changes in southern Africa, coupled with the predominant woody and perennial plants that in general can form interfertile hybrids, hybrid speciation was likely important in the flora of southern Africa.

ITS loci are generally assumed to be homogenized by concerted evolution, but factors such as hybridization, polyploidization, pseudo