| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
Systematics and Phytogeography |
2Molecular Ecology and Systematics Group, Department of Botany, Rhodes University, P.O. Box 94, Grahamstown, 6140, South Africa; 3Department of Botany, Stockholm University, Stockholm, Sweden
Received for publication October 3, 2005. Accepted for publication May 23, 2006.
ABSTRACT
Asteraceae are the largest family in southern Africa. Elucidating its origins and radiation in the region requires well-supported species-level phylogenies of the lineages. This paper presents a phylogenetic framework for subtribe Arctotidinae, which have a southern and eastern African–Australian distribution centered in the winter-rainfall region of South Africa. DNA sequence data from five chloroplast fragments (ndhF, psbA-trnH, rps16, trnS-trnfM, and trnT-trnF) and the nuclear ITS region were analyzed separately and in combination using parsimony and Bayesian methods. The data sets comprised exemplars from 18 ingroup species, representing the five currently accepted genera, and four outgroup species from Gorteriinae. All analyses indicated Arctotis and Haplocarpha are polyphyletic as presently circumscribed. The Australian-endemic Cymbonotus lawsonianus was placed within a strongly supported clade also containing A. arctotoides from South Africa and H. schimperi from eastern Africa. Retention of Dymondia and resurrection of Landtia at generic level are strongly supported. The phylogenetic hypotheses indicate the subtribe might have originated in temperate southern or eastern Africa, or it was ancestrally widespread in southern Africa and has diversified vicariously. The derived placement of C. lawsonianus indicates long-distance dispersal from southern Africa to Australia occurred.
Key Words: Arctotheca • Arctotis • chloroplast DNA • Compositae • Cymbonotus • Dymondia • Haplocarpha • ITS
Asteraceae are the largest family of angiosperms with an estimated 25 000 species in 1600–1700 genera (Bremer, 1994
). The family is well defined based on morphological and molecular data and has undergone extensive diversification. Members range from annual or perennial herbs to shrubs, trees, or vines, and extend from maritime habitats to high-alpine altitudes (Funk et al., 2005
). Asteraceae are also the largest family in southern Africa, with 253 genera and over 2250 species (Koekemoer, 1996
). Of these genera, 18% contain more than 11 species, and 6% contain more than 50. These large genera are often taxonomically difficult, and for some, adequately collected herbarium material for taxonomic revision is not available, nor are they well collected from large parts of the region, especially the more arid areas (Robertson and Barker, 2006
). Many species are highly endemic, and the semi-arid succulent Karoo biome, the Fynbos biome, and montane grasslands of the Drakensberg Mountains have high levels of endemism. Koekemoer (1996)
recognizes eight major centers of daisy endemism in southern Africa, the composition of which at tribal level is variable. Although forming a considerable component of the vegetation of the region, the origins of Asteraceae in the southern African flora remain a matter of debate and await well-supported species-level phylogenies of the genera and tribes found in the region (Galley and Linder, 2006
). Owing to the possibility of widespread reticulation and convergence in morphological traits in Asteraceae, phylogenetic hypotheses based on molecular data are an important means of resolving evolutionary and taxonomic relationships in the family. Application of molecular phylogenetic approaches has revolutionized hypotheses of evolution within Asteraceae at all taxonomic levels (e.g., Bayer and Starr, 1998
; Bayer et al., 2002
; Panero and Funk, 2002
; Funk et al., 2004
, 2005
; Barker et al., 2005
; Kim et al., 2005
). In this paper, we focus on providing the beginnings of such a phylogenetic framework for the subtribe Arctotidinae of the almost purely African tribe Arctotideae. This tribe comprises about 16 genera and 200 species (Bremer, 1994
). Most of the species occur in southern and eastern Africa, but three species are endemic to Australia.
Cassini (1816)
published the Arctotideae and subsequently distinguished two sections: Archetypae and Gorterieae (or Prototypae) (Cassini, 1823
, 1830
). Lessing (1831
, 1832
) reduced Arctotideae to a subtribe of Cynareae. Bentham (1873a
, b
) reinstated Arctotideae at tribal level and recognized three subtribes: Euarctoteae (now Arctotidinae, corresponding to Cassini's Archetypae), Gorterieae (Gorteriinae), and Gundelieae (Gundeliinae). Molecular evidence indicates that Gundeliinae is sister to Cichorieae (Karis et al., 2001
; Funk et al., 2004
) and provides support for recognition of the tribe Gundelieae (Panero and Funk, 2002
). Molecular evidence also indicates that Eremothamnus O.Hoffm., Heterolepis Cass. and Hoplophyllum DC. might belong in Arctotideae, but not in Arctotidinae or Gorteriinae (Funk et al., 2004
). However, morphological evidence for the affinities of these genera is less conclusive (Robinson and Brettell, 1973
; Reese, 1989
; Bremer, 1994
; Robinson, 1994
). Morphological and molecular evidence indicates Eremothamnus and Hoplophyllum are sister taxa (Karis, 1992
; Funk et al., 2004
). Leins (1970)
placed Eremothamnus in Arctotideae in subtribe Eremothamninae based on pollen characters, whereas other authors recognize the tribe Eremothamneae (Robinson and Brettel, 1973; Robinson, 1994
; Funk et al., 2005
).
Leaving the "problematic" genera above aside, currently Arctotideae is divided into the two subtribes Arctotidinae and Gorteriinae, of which the latter contains little more than half of the species (Bremer, 1994
). Both subtribes are supported by molecular and morphological evidence (Bremer, 1994
; Funk et al., 2004
; Karis, 2006
).
Arctotidinae are characterized by radiate heads, often shallowly lobed disc florets, obtuse inner involucral bracts with an apical scarious lamina, very small patent, subulate style sweeping hairs, as well as ribbed or grooved and often hairy cypselae with a scaly pappus. The anthers also provide a set of distinctive characters, such as the lack of tails, a rounded, short, soft and wrinkled apical appendage that contrasts with the firm, more or less sclerified, often acute, longer appendages in the rest of the subfamily. In addition, the filament collar is inconspicuous (Karis, 2006
), whereas it is very conspicuous in virtually all other taxa throughout the family, and the endothecium is "radial," i.e., the rib-like endothecial cell wall thickenings are restricted to the inner anticlinal walls of the cells (Vincent and Getliffe, 1988
). At least the latter two features and the style sweeping hairs are not found elsewhere in the subfamily, but soft anther apical appendages occur also in Cichorieae, Eremothamneae, and Heterolepis.
As presently conceived, Arctotidinae comprise approximately 80–90 species in five genera (Norlindh, 1977
; Bremer, 1994
). Arctotis L. is the largest genus with an estimated 60–70 species, all of which are indigenous to southern Africa with the exception of a single species, A. maidenii Beauverd, which has a restricted distribution in eastern Australia. Arctotheca J.C. Wendl. contains five species indigenous to southern Africa (Germishuizen and Meyer, 2003
). Cymbonotus Cass. comprises two species endemic to Australia. Murray (1992)
placed A. maidenii in Cymbonotus (as C. sp. ‘A') but did not make the formal nomenclatural transfer. Dymondia Compton is monotypic and confined to the Agulhas Plains in South Africa (Compton, 1953
). Haplocarpha comprises nine species, of which six species are found in southern Africa and three species occur in afromontane eastern Africa as far north as Eritrea (Hedberg, 1957
; Pope, 1992
; Germishuizen and Meyer, 2003
).
Delimitation of genera in Arctotidinae has always been controversial. Instability in generic classification in Arctotidinae partly reflects differences in the morphological characteristics emphasized to delimit genera by different authors. For example, Lessing (1832)
emphasized fruit and pappus morphology, ray floret fertility, and filament ornamentation. However, fruit pubescence and pappus characters were found to be unreliable generic criteria in the subtribe (Beauverd, 1915
; Lewin, 1922a
). Regarding the five currently accepted genera, Arctotis is characterized by the possession of three well-developed abaxial wings on the cypsela, that create one or two distinct furrows or "cavities" (McKenzie et al., 2005
). Cymbonotus reportedly differs in possessing "somewhat papillose" filaments, glabrous cypselae with two lateral and two abaxial ribs, and lacking a pappus (Bremer, 1994
). However, in cypselar morphology Cymbonotus is barely distinguishable from species placed in Arctotis sect. Austro-orientales K.Lewin (McKenzie et al., 2005
) and at least some Cymbonotus specimens have smooth filaments (Muller, 2003
). In Arctotheca and Haplocarpha the cypselae lack distinct wings and instead have longitudinal ridges. Arctotheca and Haplocarpha are discriminated by the presence of neuter ray florets in Arctotheca and female ray florets in Haplocarpha. Based on cypselar morphology, Haplocarpha comprises three disparate species groups differing in pappus morphology, cypsela pubescence, width of the longitudinal ridges on the cypsela and presence of a transversely rugose or smooth cypsela surface (McKenzie et al., 2005
). Dymondia is characterized by the combination of a rhizomatous habit, sessile capitula, non-appendiculate involucral bracts, densely hairy alveolate receptacle, and biseriate pappus scales (Compton, 1953
). Mature cypselae have never been seen in D. margaretae Compton.
Four additional genera were accepted by Harvey (1865)
. Venidium Less. was segregated from Arctotis based on the presence of less well-developed cypsela wings, the absence or extreme reduction of pappus scales, and differences in cypsela pubescence (Lessing, 1831
, 1832
). Venidium comprised species that Lewin (1922a)
placed in Arctotis sect. Acuminatae K.Lewin, Austro-orientales, Caudatae K.Lewin, and Hirsutae K.Lewin. Cypsela pubescence and pappus characters were used to discriminate Cryptostemma R.Br. and Microstephium Less. from Arctotheca. Landtia was discriminated from Haplocarpha primarily by the presence of smooth filaments (Lessing, 1832
), but H. parvifolia (Schltr.) Beauverd also possesses smooth filaments (Lewin, 1922a
). Other characters distinguishing Landtia and Haplocarpha have been proposed, such as pappus scale apex shape, cypsela shape, rugose versus smooth cypsela surface, and pappus length relative to cypsela length (e.g., Harvey, 1865
; Bentham, 1873a
; Phillips, 1951
). However, the features used to segregate these four genera were subsequently concluded to be untenable (Beauverd, 1915
; Lewin, 1922a
).
Ongoing morphological and molecular studies indicate that the current taxonomy of Arctotidinae is in need of further resolution and current generic concepts do not satisfactorily delimit "natural" groups. DNA sequence data have been utilized widely in Asteraceae to help elucidate generic limits and relationships, but have tended to exploit a narrow range of DNA regions, especially the ndhF, rbcL, psbA-trnH, and trnL-trnF chloroplast regions and the internal transcribed spacers (ITS) region from nuclear ribosomal DNA (nrDNA). Previous molecular studies that have incorporated Arctotidinae were focused on tribal and subtribal relationships (Kim and Jansen, 1995
; Karis et al., 2001
; Funk et al., 2004
) and included insufficient samples to rigorously test the existing morphology-based generic classification. At present, sequence data are known for seven species of Arctotidinae for the ndhF gene, trnL intron, trnL-trnF intergenic spacer, and the ITS region (Kim and Jansen, 1995
; Funk et al., 2004
). In addition, three genes (trnE, trnG, and trnY) and three intergenic spacers (trnS-trnC, trnT-trnG, and trnY-rpoB) have been sequenced for Arctotis stoechadifolia P.J. Bergius (Kim et al., 2005
). Primers have been developed for an ever-increasing number of DNA regions and comparative studies indicate that the most commonly utilized regions (such as trnL-trnF) are not always the most variable or informative in a given taxonomic group (Shaw et al., 2005
). As highlighted in recent reviews (Álvarez and Wendel, 2003
; Bailey et al., 2003
; Small et al., 2004
), caution is required in the phylogenetic interpretation of ITS sequence data. The use of ITS is complicated by potential problems with orthology and paralogy, the presence of pseudogenes, concerted evolution, hybridization and lineage sorting of ancestral polymorphisms, which may give rise to misleading phylogenies (Álvarez and Wendel, 2003
). Given these caveats, it is highly desirable that phylogenetic inference derived from ITS sequences is corroborated with independent sources of evidence, such as cpDNA and morphological data. Strong congruence between independent data sets will raise the level of confidence in the phylogenetic hypotheses.
Our ultimate objective is to produce a well-resolved species-level phylogeny for Arctotidinae, with emphasis on the genus Arctotis, which can be utilized to investigate the biogeography and evolution of Arctotidinae. A prerequisite to achieving this is to resolve generic delimitation in the subtribe. The aim of this paper is to generate a phylogenetic hypothesis for Arctotidinae based on cpDNA and nrDNA sequences. The phylogeny will provide a sound foundation for resolution of generic concepts in conjunction with other types of evidence.
MATERIALS AND METHODS
Taxon sampling
Twenty-two species representing nine genera were included in the study (see Appendix). The ingroup comprised single exemplars of 18 species representing the five genera of Arctotidinae. The ingroup species were selected to encompass the main morphological diversity in Arctotidinae. Single accessions from each of four genera of Gorteriinae were selected as the outgroup. Generic concepts follow Bremer (1994)
. Lewin's (1922a) infrageneric classification of Arctotis and Beauverd's (1915) infrageneric classification of Haplocarpha are followed. The South African species recognized mainly follow Germishuizen and Meyer (2003)
. The specimen N. P. Barker 1865 belongs to the undetermined Arctotis sp. ‘2' designated by J. B. P. Beyers (in Goldblatt and Manning, 2000
). This species is referred to as ‘Arctotis sp.' herein.
DNA extraction, amplification, and sequencing
Total genomic DNA was extracted from about 1 cm2 of either fresh leaf material or leaves desiccated in silica gel using a CTAB extraction protocol (Doyle and Doyle, 1987
). Five cpDNA fragments (ndhF gene, psbA-trnH intergenic spacer, rps16 intron, and the trnS-trnfM and trnT-trnF regions) and one nrDNA region (ITS) were amplified and sequenced. The trnL intron and the neighboring trnT-trnL and trnL-trnF intergenic spacers (collectively termed the trnT-trnF region herein) are among the most commonly utilized cpDNA regions in plant phylogenetics from species to family levels (Sang et al., 1997
; Bayer and Starr, 1998
; Borsch et al., 2003
; Shaw et al., 2005
; Won and Renner, 2005
). The rps16 intron and psbA-trnH intergenic spacer are considered to be relatively fast-evolving cpDNA regions and have been utilized from family to intrageneric levels (Oxelman et al., 1997
; Sang et al., 1997
; Asmussen and Chase, 2001
; Mort et al., 2002
; Hamilton et al., 2003
). The trnS-trnfM region, which includes the psbZ and trnG coding regions (Shaw et al., 2005
), has not been used widely in plant phylogenetics, but has proved informative at the intraspecific level in Eritrichium nanum (L.) Gaudin (Stehlik et al., 2002
) and interspecific level in Citrullus Schrad. ex Eckl. & Zeyh. (Dane et al., 2004
). The ndhF gene is one of the most rapidly evolving cpDNA coding regions (Olmstead and Palmer, 1994
). However, it typically evolves at a slower rate than noncoding regions and therefore is mainly utilized at suprageneric levels (Olmstead and Palmer, 1994
; Kim and Jansen, 1995
). Sequence data from the ITS nrDNA region, which comprises the 5.8S gene and the flanking internal transcribed spacers ITS1 and ITS2 (Baldwin, 1992
), have been utilized widely in the Asteraceae for inferring phylogenetic relationships (e.g., Baldwin, 1992
; Barker et al., 2005
and references therein).
The primers used to amplify and sequence these regions are listed in Appendix S1 (see Supplemental Data accompanying online version of this article). The trnT-trnF region was amplified and sequenced in two overlapping fragments, using the primer pairs A/D and C/F. For some samples, the trnT-trnL and trnL-trnF spacers were amplified separately with the A/B and E/F primer pairs, respectively. The ndhF gene was amplified in overlapping fragments with the primer pairs ndhF-5'F/ndhF-1074R and ndhF-913F/ndhF-3'R. Occasionally, the internal ndhF primers were used to amplify the region in smaller fragments and for sequencing. The ITS region was amplified with the external primer pairs ITS1/ITS4, ITS5/ITS4 or ITS18/ITS26; the internal primers Chromo-5.8R and Chrysanth-5.8F were used for sequencing. The internal primers trnS-fM-478F and trnS-fM-582R were used to amplify the trnS-trnfM region in two fragments for Haplocarpha rueppellii (Sch. Bip.) Beauverd.
For polymerase chain reaction (PCR) amplification of all cpDNA regions except trnS-trnfM, each 25 µL reaction solution contained 2.5 µL 10x PCR reaction buffer (Bioline, London, England), 1 µL 20 mM dNTPs (Bioline), 1 µL 0.1 µM solution of each forward and reverse primer, 0.2 µL BioTaq DNA polymerase (5 units/µL, Bioline) and 0.75–1.5 µL unquantified DNA extract. The volume of 50 mM MgCl2 varied from 0.75 µL (1.5 mM final concentration) for rps16, 0.75–1.5 µL (1.5–3 mM) for ndhF and trnT-trnF, and 2.5–3 µL (5–6 mM) for psbA-trnH. Some reaction solutions contained 1.5–2 µL 0.1% bovine serum albumen. The 25 µL amplification reaction solution for the trnS-trnfM and ITS regions differed in containing 5 µL 5x Colorless GoTaq reaction buffer, 0.25 µL GoTaq DNA polymerase (5 units/µL, Promega, Madison, Wisconsin, USA) and no additional MgCl2.
The DNA regions were amplified using a Hybaid (Ashford, England) PCR Sprint thermal cycler. The following parameters in the amplification reactions were standard for all regions: in the first reaction cycle, denaturing 95°C, 45 s; primer extension 72°C, 3 min; and the final extension cycle 72°C, 10 min. The number of amplification cycles and primer annealing temperature varied as follows: ndhF—30 cycles, 50–52°C, 45 s; rps16 and trnS-trnfM—33–35 cycles, 52°C, 45 s; psbA-trnH—25 cycles, 52°C, 45 s; trnT-trnL spacer (A and B)—27–30 cycles, 52°C, 45 s; trnT-trnL spacer and trnL intron (A and D)—35 cycles, 52°C, 45 s; trnL-trnF spacer and trnL intron (C and F): 30 cycles, 52°C, 45 s; trnL-trnF spacer (E and F): 25–30 cycles, 52°C, 45 s; ITS: 30–33 cycles, 48–52°C, 45 s.
The PCR products were purified using the Wizard SV Gel and PCR purification kit (Promega) and resuspended in 25–30 µL water. Owing to the presence of double bands in the PCR products amplified with the C and F primers for Arctotis dregei Turcz., the PCR products were separated on a 1% agarose gel, the bands excised and purified using the Wizard kit according to the manufacturer's instructions.
Sequencing reactions of the PCR products were performed using the BigDye Terminator, v. 3.1, cycle sequencing kit (Applied Biosystems, Foster City, California, USA). The cycle sequencing protocol was: denaturing 96°C, 30 s; primer annealing 50°C, 15 s; primer extension 60°C, 4 min. The cycle-sequencing products were precipitated using the sodium acetate/EDTA protocol, and electrophoresed and resolved using an ABI Prism 3100 Genetic Analyzer (Applied Biosystems).
Sequence alignment
For each accession, contiguous sequences were compiled with Sequencher 4.2.2 (Gene Codes, Ann Arbor, Michigan, USA) and edited visually. All sequences were deposited in GenBank (accession numbers listed in Appendix). Sequences were aligned manually in MacClade 4.06 (Maddison and Maddison, 2000
). Gaps were inserted manually based on visual inspection of the sequences with the aim of minimizing the number of gaps inserted. For parsimony analyses, unambiguously aligned insertion/deletion events (indels) greater than 2 bp in length were recoded as binary (presence/absence) characters, following Bayer et al. (2002)
, using the "simple indel coding" method of Simmons and Ochoterena (2000)
. Owing to the variability of the ITS sequences, no unambiguously aligned indels longer than 2 bp were recoded for the ITS data set.
Phylogenetic analyses
Because the chloroplast genome is inherited as a single unit without recombination, combining sequences from multiple cpDNA regions is justified (Soltis and Soltis, 1998
). Furthermore, improved phylogenetic resolution in analyses of combined cpDNA data sets, compared to separate analyses of the individual regions, is often reported in the literature (e.g., Asmussen and Chase, 2001
; Kress et al., 2001
). Thus three data sets were analyzed: the combined cpDNA data set, the ITS data set, and a total evidence approach (Kluge, 1989
; Nixon and Carpenter, 1996
) with the combined cpDNA and ITS data sets. Congruence in the phylogenetic signal of the cpDNA and ITS data sets was examined by a visual comparison of tree topologies and branch support, and conducting a partition homogeneity (or ILD) test (Farris et al., 1994
) with PAUP* 4.0b10 (Swofford, 2002
) using a heuristic search with 1000 replicates, a maximum tree limit of 1000, and tree-bisection-reconnection (TBR) branch swapping.
Parsimony analyses were performed with PAUP*. For each data set, a heuristic search was conducted with 1000 simple taxon-addition replicates, TBR branch swapping, and the MULTREES option in effect. Further heuristic searches for equally most parsimonious trees (MPTs) employing 1000 random taxon-addition replicates, holding 10 trees at each step, TBR branch swapping, with the MULTREES option in effect, and saving a maximum of 30 trees per replicate, did not recover shorter MPTs. Uninformative characters were excluded before all analyses, and all nucleotide characters were equally weighted. The consistency index (CI; Kluge and Farris, 1969
) and retention index (RI; Farris, 1989
) were calculated as estimates of how well the data fitted the inferred phylogenies. Data sets were analyzed with and without recoded indels. Analysis of the ITS data set was repeated with seven ambiguously aligned regions (bp 80–98, 124–134, 161–163, 236–238, 265–276, 483–509, and 637–651) excluded. Relative branch support was assessed using bootstrap (BS) resampling (Felsenstein, 1985
) with 1000 replicates, holding 10 trees at each step, simple addition sequence, TBR branch swapping, and MULTREES and STEEPEST DECENT options in effect.
Sequences from the chloroplast genomes of Lactuca sativa L., Nicotiana sylvestris Speg., and N. tabacum L. from GenBank (http://www.ncbi.nlm.nih.gov) and the Asteraceae ITS secondary structure model of Goertzen et al. (2003)
were used as guides to partition sequences into coding, exon, intron, and intergenic spacer regions prior to Bayesian inference (BI) analyses. Sixteen partitions (13 cpDNA, three ITS) were defined (Appendix S2, see Supplemental Data accompanying online version of this article). The ndhF sequence included the ndhF gene and ndhF–rpl32 intergenic spacer. The psbA-trnH sequence comprised the psbA-trnH spacer, 53 bp at the 3' end of the psbA gene, and 27 bp at the 5' end of the trnH sequence. The rps16 sequence comprised the intron only. The trnS-trnfM region comprised the psbZ and trnG coding regions and the trnS-psbZ, psbZ-trnG, and trnG-trnfM spacers (see Shaw et al., 2005
). The trnT-trnF region comprised the trnL intron, the flanking trnL 5' and 3' exons, and the trnT-trnL and trnL-trnF intergenic spacers. The ITS region comprised the 5.8S gene and the flanking internal transcribed spacers ITS1 and ITS2 (Baldwin, 1992
). Owing to the minimal variability of the trnL exons (two variable sites and one parsimony-informative site) and partial trnH sequence (one variable site), they were deleted from the data set for all combined-data analyses. An optimal nucleotide-substitution model for each partition (Appendix S2, see Supplemental Data accompanying online version of this article) was selected using the Akaike information criterion with MrModeltest 2.2 (J. A. A. Nylander, Uppsala University, Sweden).
BI using Markov chain Monte Carlo methods (Yang and Rannala, 1997
) was performed using MrBayes 3.1.1 (Huelsenbeck and Ronquist, 2001
). Two independent analyses each with four Markov chains, three heated and one cold, starting from a random tree were run simultaneously for 1 x 106 generations with trees sampled every 100 generations. The trees sampled prior to stabilization of the log-likelihood value (the first 10 000 generations in each analysis) were discarded as "burn-in" samples. As a measure of support for the sampled trees containing a particular clade, posterior probabilities (PP) were estimated by constructing a 50% majority rule consensus tree in PAUP*.
RESULTS
The sequence characteristics for each DNA fragment and the combined data sets are summarized in Table 1. The following strongly supported clades were retrieved in all analyses. Arctotidinae species formed a highly supported clade (BS = 100%; PP = 1, Figs. 1, 2). In the ingroup, four clades were consistently retrieved: (1) a ‘Landtia' clade comprising two species from Haplocarpha subg. Landtia (H. nervosa (Thunb.) Beauverd and H. rueppellii), which in all analyses was sister to the rest of the Arctotidinae (BS = 100%; PP = 1); (2) an ‘Arctotis' clade comprising Arctotis acaulis L., A. aspera L., Arctotis sp. and A. venusta Norl. (BS 
84%; PP = 1); (3) a ‘Cymbonotus' clade containing Cymbonotus lawsonianus Gaudich., Arctotis arctotoides (L.f.) O. Hoffm. and Haplocarpha schimperi (Sch. Bip.) Beauverd (BS 84%; PP = 1); (4) a clade comprising the sampled members of Arctotis sect. Anomalae (A. dregei and A. sulcocarpa K.Lewin) (BS 94%; PP = 1). Dymondia margaretae and Haplocarpha scaposa Harv. were always placed on monotypic branches.
|
|
|
The combined cpDNA data set comprised 6370 bp of which 562 sites (8.8%) were variable and 262 sites (4.1%) were parsimony informative. Parsimony analysis yielded a single MPT of 367 steps (CI = 0.793, RI = 0.883). The 50% majority rule consensus tree from a BI analysis (Fig. 1) differed from the MPT in the placement of A. breviscapa (sister to a ‘Cymbonotus'–Arctotheca–‘core Haplocarpha' [comprising H. lanata and H. lyrata]–Arctotis sect. Anomalae clade [BS < 50%] in the MPT, and sister to the ‘Arctotis' clade [PP = 0.83] in the BI phylogeny), and two outgroup species, Berkheya carduoides (Less.) Hutch. and Cuspidia cernua (L.f.) B.L. Burtt, formed a clade (BS = 100%) in the MPT. In addition to the placements common to all analyses already mentioned, Arctotis perfoliata (L.f.) O. Hoffm. was sister to the ‘Cymbonotus' clade (BS = 89%, PP = 1). Arctotheca calendula (L.) Levyns was placed sister to H. lyrata Harv. (BS = 100%, PP = 1) within a clade also including H. lanata (BS = 99%, PP = 1). The Arctotis sect. Anomalae clade was sister to the Arctotheca–‘core Haplocarpha' clade (BS = 91%, PP = 1). In the MPT the strongly supported ‘Arctotis' clade was sister to a large clade containing Arctotheca calendula, Arctotis sect. Anomalae, A. breviscapa, A. perfoliata, the ‘Cymbonotus' clade, and the ‘core Haplocarpha' clade (BS = 89%).
Recoding indels as binary characters had no impact on tree topology, and BS was affected notably for the Arctotis acaulis–A. venusta node only (BS increased from 62 to 87%).
Internal transcribed spacers data set
The ITS sequences were considerably more variable than any of the individual cpDNA fragments and contained 251 variable sites (34.2%) and 176 parsimony-informative sites (24%) (Table 1). Parsimony analysis of the complete data set yielded two MPTs of 434 steps (CI = 0.668, RI = 0.735). The two MPTs differed in that the Arctotis sect. Anomalae clade was placed either sister to ‘core Haplocarpha' or an ‘Arctotis'–A. breviscapa–A. perfoliata clade. The 50% majority rule consensus tree obtained from the BI analysis was almost identical in topology (Fig. 1). The only difference was that Arctotheca calendula was placed in a terminal clade with Haplocarpha lanata and H. lyrata in the BI tree (PP = 0.97), but was sister to a large Arctotis–‘core Haplocarpha'–‘Cymbonotus' clade in the strict consensus tree from the parsimony analysis (BS < 50%). The basal nodes within this large clade were poorly supported (BS 
56%, PP = 0.91 and 0.95). Arctotis perfoliata was sister to the ‘Arctotis' clade but with poor support in both analyses (BS = 58%, PP = 0.64). Arctotis breviscapa was sister to this ‘Arctotis'–A. perfoliata clade (BS < 50%, PP = 0.98).
Exclusion of seven ambiguously aligned regions reduced the data set to 643 characters, of which 129 were parsimony informative. Parsimony analysis of the reduced data set yielded 60 MPTs of 309 steps (CI = 0.663, RI = 0.736). The major clades retrieved were identical to those obtained from analysis of the complete ITS data set. However, the strict consensus tree topology differed in that the Arctotheca calendula, ‘Arctotis', ‘Cymbonotus', Arctotis breviscapa, Arctotis sect. Anomalae, Arctotis perfoliata and ‘core Haplocarpha' lineages formed a polytomy.
Congruence assessment
Inspection of tree topologies revealed that the placement and branch support for Arctotis perfoliata conflicted strongly between the cpDNA and ITS phylogenies. Arctotis perfoliata was sister to the ‘Cymbonotus' clade in the cpDNA analyses (BS = 89%, PP = 1) and sister to the ‘Arctotis' clade in the ITS analyses but with poor support (BS = 58%, PP = 0.64). The placement of Arctotheca calendula and Arctotis breviscapa showed variation with the method of data analysis. Arctotheca calendula was placed within a strongly supported terminal clade with Haplocarpha lanata and H. lyrata (BS = 100%, PP = 1) in parsimony and BI analyses of the cpDNA data. BI analysis of the ITS data supported this placement (PP = 0.97), whereas parsimony placed it sister to the main ‘Arctotis'–‘core Haplocarpha'–‘Cymbonotus' clade but with poor support (BS < 50%). Arctotis breviscapa was sister to the ‘Arctotis' clade (BS < 50%, PP = 0.98) in both BI and parsimony analyses of the ITS data set, and BI analysis of the cpDNA data set (PP = 0.83). Parsimony analysis of the cpDNA data set placed A. breviscapa sister to a large clade containing A. calendula, ‘core Haplocarpha', Arctotis sect. Anomalae and the ‘Cymbonotus' clade with poor support (BS < 50%). The placement of three other species (Arctotis venusta, H. lanata and H. lyrata) differed between the cpDNA and ITS phylogenies, but only within the same clade. For both Haplocarpha species, the nodes in both phylogenies were strongly supported (BS 91%, PP = 1). The placement of A. venusta received weaker support (e.g., BS = 62%, PP = 0.97 in the cpDNA phylogeny).
An ILD test (
= 0.05) found significant heterogeneity between the cpDNA and ITS data sets (P = 0.003). Sequential exclusion of the six species with variable phylogenetic placement (A. calendula, A. breviscapa, A. perfoliata, A. venusta, H. lanata, and H. lyrata) markedly improved congruence only when H. lanata and H. lyrata were excluded (P = 0.080 and 0.091, respectively). In pairwise exclusions involving the other four species, the data sets were highly congruent only when A. calendula and A. perfoliata (P = 0.529), and A. calendula and A. venusta (P = 0.141) were excluded.
Total evidence
Parsimony analysis yielded a single MPT of 788 steps (CI = 0.711, RI = 0.809). The 50% majority rule consensus tree from a BI analysis (Fig. 2) was identical in topology, except that two of the outgroup species, Berkheya carduoides and Cuspidia cernua, were placed on separate branches. The tree topology was identical to those from the cpDNA data analyses, except that Arctotis venusta was placed as sister to the remainder of the ‘Arctotis' clade. Nodes were strongly supported, except the interior branches linking A. breviscapa, A. perfoliata, the ‘Cymbonotus' clade, and a strongly supported clade comprising Arctotheca calendula, ‘core Haplocarpha' and Arctotis sect. Anomalae.
Sequential exclusion from the data set of the six species with variable phylogenetic placement had no impact on the major clades retrieved; only the placement of other "problematic" species varied. Exclusion of A. breviscapa yielded the shortest MPT (760 steps, CI = 0.734, RI = 0.827). When A. perfoliata was excluded, two MPTs (772 steps, CI = 0.725, RI = 0.820) were obtained, in which A. breviscapa was sister to either the ‘Arctotis' clade or a ‘Cymbonotus'–Arctotis sect. Anomalae–‘core Haplocarpha'–Arctotheca clade.
DISCUSSION
Congruence between the cpDNA and ITS data sets
Inspection of tree topologies and branch support values indicated a major source of conflict between the cpDNA and ITS data sets was the placement of Arctotheca calendula and Arctotis perfoliata. Exclusion of these species from analyses had little impact on the rest of the topology, indicating the major clades are robust and reflect strong phylogenetic signal in the data sets. Topological differences can be due to weakly supported clades resulting from insufficient phylogenetic signal in the separate data sets, or different clades may be strongly supported in different topologies due to separate evolutionary histories. The latter is a distinct possibility in plants when utilizing sequence data from different genomes (e.g., Mason-Gamer and Kellogg, 1996
). No single congruence test has received universal approval and each has limitations (see Johnson and Soltis, 1998
). Mason-Gamer and Kellogg (1996)
evaluated methods of assessing conflict between data sets and concluded combination of data sets is possible if the trees do not conflict or if conflicting nodes receive low BS support. Nodes receiving BS > 70% are interpreted as incongruent. In the present study, BS and PP support for the placement of A. calendula and A. perfoliata in analyses of the cpDNA data set was high (BS 89%, PP = 1), but support was weaker for their placement in the ITS data analyses (BS < 58%, PP = 0.97 and 0.62, respectively).
The ILD test indicated incongruence existed between the cpDNA and ITS data sets. However, significant P values are not necessarily indicative of phylogenetic incongruence between independent data sets (Yoder et al., 2001
; Hipp et al., 2004
and references therein). The ILD test can be sensitive to small differences in topology and character congruence, and does not reveal the extent of incongruence (Siddall, 1997
). Tests with empirical data have found that P values as low as 0.001 should not prohibit combination of data sets (see Yoder et al., 2001
). In the present study, the degree of incongruence indicated by the ILD test partly reflected the level of support for the conflicting nodes and the "decisiveness" (Goloboff, 1991
) of the data sets. The ILD test indicated Haplocarpha lanata and H. lyrata to be the greatest contributors to incongruence, followed by A. calendula and A. perfoliata. Although the placement of the two Haplocarpha species only varied within the same clade, the nodes were strongly supported. Arctotis breviscapa was placed sister to different clades by the different methods of analyzing the cpDNA data set, but these nodes were poorly supported and the ILD test indicated A. breviscapa was not a major contributor to incongruence. Application of criteria supported by Mason-Gamer and Kellogg (1996)
would also lead to the conclusion that A. breviscapa was not the major source of data set conflict. In contrast, the placement of A. calendula and A. perfoliata was strongly supported in the cpDNA phylogeny, but weakly supported in the ITS tree. The placement of A. calendula from ITS sequences differed between parsimony and BI analyses, reflecting differences in the analytic methods employed. Arctotheca calendula was linked with Haplocarpha lanata and H. lyrata in the BI analysis (PP = 0.97), but this branch received no bootstrap support and the interior branches of the MPTs derived from the ITS data set were generally weakly supported (BS 58%, PP 0.91). The order of divergence of the terminal clades also differed somewhat between the cpDNA and ITS data sets, but the internal nodes are indicated to be based on relatively few characters and were weakly supported.
A single well-resolved MPT with high BS, and a BI phylogeny with high PP, for most nodes was obtained from analyses of the combined cpDNA data set and the total evidence. One interpretation of this outcome is that the data sets are largely congruent and supports their combination. However, as highlighted by Soltis et al. (2004)
, caution is needed when interpreting such results. Inadequate taxon sampling (using too few or unrepresentative taxa) may generate a single well-supported, but incorrect, phylogeny. Individual taxa can be greatly misplaced in analyses of small taxon samples (Soltis et al., 2004
). Our ingroup sample covers the main morphological diversity present in Arctotidinae, therefore we believe recovery of incorrect topologies and other artefacts (e.g., masking of paraphyly) due to inadequate taxon sampling is unlikely to be a major factor in the present study.
The close correspondence of the total evidence tree topology with that of the cpDNA phylogeny might simply reflect the preponderance of parsimony-informative cpDNA sites compared to ITS sites (262 vs. 176, respectively), and the shorter cpDNA MPT compared to the ITS MPTs (367 vs. 434 steps, respectively), rather than a less decisive phylogenetic signal (sensu Goloboff, 1991
) in the ITS data set.
The higher PP values in BI analyses compared to BS values obtained in our investigation is consistent with other studies (e.g., Suzuki et al., 2002
; Wilcox et al., 2002
; Haston et al., 2005
). This at least partly reflects differences in the tree sampling methods of the two support measures. However, some simulation studies have found PP values to be excessive overestimates of support, whereas bootstrap and jackknife approaches are more conservative estimates of confidence (Suzuki et al., 2002
; Simmons et al., 2004
).
Taxonomic implications
The concurrence of the phylogenies with morphology can be assessed based on published descriptions (Lessing, 1831
, 1832
; Harvey, 1865
; Bentham 1873a
, b
; Beauverd, 1915
; Lewin, 1922a
; Murray, 1992
; Bremer, 1994
; McKenzie et al., 2005
) and personal observations. Morphological evidence supports the phylogenetic hypotheses in indicating that generic concepts in Arctotidinae are in need of revision.
Arctotheca
Only one Arctotheca species, A. calendula, was included in the present study. The cpDNA data, and BI analysis of the ITS data set, strongly supported a close phylogenetic relationship between A. calendula, Haplocarpha lanata, and H. lyrata. These three species have hermaphrodite disc florets, papillose filaments, and lack abaxial wings on the cypselae, but differ in many morphological features, including ray floret fertility, pappus length relative to cypsela length, and cypsela shape and pubescence. Sampling of other Arctotheca species is required to clarify the phylogenetic relationships of Arctotheca calendula and investigate the delimitation of Arctotheca. On the basis of morphology, Arctotheca is well defined and characterized by neuter ray florets, hermaphrodite disc florets, papillose filaments, bilaterally flattened cypselae with well-developed abaxial ridges but lacking wings, and a pappus never longer than the cypsela (Lewin, 1922a
; McKenzie et al., 2005
).
Arctotis
The present study highlights the polyphyly of Arctotis as the genus is defined currently and indicates that Arctotis species belong to at least three distinct clades. The placement of the sampled Arctotis species in different clades is well supported by morphological evidence.
The type species of Arctotis is A. angustifolia L. In morphology A. angustifolia is very similar to the four species comprising the ‘Arctotis' clade (i.e., A. acaulis, A. aspera, A. sp., and A. venusta). In these species, the ray florets are female, the outermost disc florets are hermaphrodite, and the innermost disc florets are either male or sterile. The filaments are smooth. The cypselae have three well-developed abaxial wings, that create two distinct cavities, and have a well-developed basal coma of twin hairs. The pappus is composed of free scales longer than the cypsela and in two distinct series (McKenzie et al., 2005
).
Some conflict in the phylogenetic placement of A. breviscapa was obtained in this study. Analyses of ITS sequences and BI analysis of the cpDNA data set indicated A. breviscapa to be an early divergence in the ‘Arctotis' lineage, whereas parsimony analysis of the cpDNA data indicated the species to be sister to an Arctotheca–Arctotis sect. Anomalae–‘Cymbonotus'–‘core Haplocarpha' clade and to have diverged after the ‘Arctotis' lineage. Lewin (1922a)
placed A. breviscapa in its own section (sect. Leptorhizae K. Lewin). The species is anomalous within Arctotis in possessing two well-developed adaxial wings, in addition to three abaxial wings, on the cypselae. Further evidence is required to resolve its evolutionary history and taxonomic placement.
Some uncertainty also exists over the phylogenetic placement of Arctotis dregei and A. sulcocarpa. Analyses of the cpDNA data set placed the two species as sister to an Arctotheca calendula–‘core Haplocarpha' (see Haplocarpha below) clade (BS = 91%, PP = 1). In contrast, analyses of ITS sequences indicate the two species formed a strongly supported clade within a polytomy also including the Arctotheca calendula–‘core Haplocarpha' clade and an ‘Arctotis'–A. perfoliata–A. breviscapa clade. Morphological evidence indicates A. dregei and A. sulcocarpa are misplaced in Arctotis. Arctotis dregei and A. sulcocarpa share a similar cypselar morphology to the ‘Arctotis' clade (presence of abaxial cypsela wings, a basal coma of twin hairs, and free pappus scales usually longer than the cypsela), but differ in having neuter ray florets, papillose filaments, and cypselae that are bilaterally flattened, have weakly developed adaxial wings and lack a well-developed median abaxial wing. In floral and vegetative morphology, A. dregei and A. sulcocarpa are extremely similar to Arctotheca calendula. These three species share an annual life history, pinnatisect leaves with a hirsute adaxial surface and densely lanate abaxial surface, neuter ray florets, hermaphrodite disc florets, and papillose filaments. Further investigations utilizing single-copy nuclear genes or molecular markers, such as nuclear microsatellite loci, would help to elucidate whether hybridization or lineage sorting are responsible for the phylogenetic placements of A. dregei and A. sulcocarpa.
Cymbonotus
The present study agrees with morphological evidence in indicating that the boundary between Cymbonotus and Arctotis requires further scrutiny. The phylogenetic hypotheses presented herein provide strong support for a clade including the Australian Cymbonotus lawsonianus, the South African Arctotis arctotoides, and the eastern African Haplocarpha schimperi. Additional molecular data (R. J. McKenzie, S. D. Mitchell, E. M. Muller, and N. P. Barker, Rhodes University, unpublished data) indicates other taxa currently placed in Arctotis sect. Austro-orientales, with the exception of A. perfoliata, also belong to this clade. In this study, the ‘Cymbonotus' clade is phylogenetically distinct from the ‘Arctotis' clade, but interior nodes of the trees are weakly supported. It is possible that with the addition of more data or taxa the ‘Cymbonotus' clade might be included within a broader ‘Arctotis' clade. The phylogenetic hypotheses presented herein do not support the reduction of Cymbonotus to a subgenus within Arctotis (Beauverd, 1915
), unless an extremely broad concept of Arctotis, encompassing the majority of the species in Arctotidinae, is envisaged. Given the morphological diversity among these species, circumscribing such a genus based on morphological traits would be difficult. Baillon (1888)
and Hoffman (1897)
defined Arctotis broadly to include all of the then-known taxa currently placed in Arctotidinae, but this was motivated by the inclusion of Ursinia Gaertn. within Arctotidinae. Ursinia is now known to belong in the Anthemideae (Karis, 1993
; Bremer, 1994
; Kim and Jansen, 1995
), but apparently was associated with Arctotideae due to possessing similar involucral bracts with a scarious lamina, and a pappus of scales.
The genus Cymbonotus was published by Cassini (1825)
and has been distinguished from Arctotis by possessing glabrous cypselae with two lateral and two abaxial ribs, the absence of a pappus, and papillose filaments (Bremer, 1994
). Lessing (1831
, 1832
) discriminated Venidium from Arctotis by the possession of glabrous cypselae devoid of a pappus. Lessing (1832)
recognized eight Venidium species, but evidently had some doubts about the distinctness of Venidium and Cymbonotus. The cypselae of the two Cymbonotus species and Arctotis sect. Austro-orientales (excluding A. perfoliata) have similarly developed abaxial wings, lack pappus scales and a basal coma of twin hairs, and are either glabrous or bear simple uniseriate or papillose trichomes. In addition, certain entities in the A. fastuosa Jacq. species complex bear similarities in cypselar morphology to A. sect. Austro-orientales (McKenzie et al., 2005
). Based on the strong similarity in cypselar morphology, Beauverd (1915)
reduced Cymbonotus and Venidium to subgenera of Arctotis. Lewin (1922a)
recognized Cymbonotus but noted that support from morphological evidence was weak and the main basis for its continued segregation was its geographical isolation. After reading Beauverd's paper, Lewin (1922b)
concurred with Beauverd's classification. Nevertheless, subsequent authors (Norlindh, 1977
; Murray, 1992
; Bremer, 1994
) have continued to recognize Cymbonotus.
The present study does not support recognition of a monophyletic Venidium. The type species of Venidium is V. perfoliatum (L.f.) Less. (now Arctotis perfoliata). This species is currently placed in Arctotis sect. Austro-orientales (Lewin, 1922a
). The cypselae of A. perfoliata lack a pappus and a basal coma, features shared with Cymbonotus and other species in sect. Austro-orientales, but bear three well-developed abaxial wings that create two distinct "cavities." The latter traits are shared with the ‘Arctotis' clade. In contrast, the cypselae of Cymbonotus and other species in A. sect. Austro-orientales have strongly inflexed lateral wings and a more weakly developed median wing, creating a single abaxial "furrow" (McKenzie et al., 2005
). Female ray florets, hermaphrodite disc florets, and smooth filaments are other features A. perfoliata shares with Cymbonotus and A. sect. Austro-orientales. Given the fundamental differences in cypselar morphology and the variable phylogenetic placement of A. perfoliata in the present study, additional evidence is required to resolve its taxonomic placement. Inclusion of a wider sample of taxa or sampling additional, more variable DNA regions might help to elucidate the evolutionary history of A. perfoliata.
Of the eight Venidium species recognized by Lessing (1832)
, four have been reduced to synonymy and four are currently recognized at species level in Arctotis: A. arctotoides, A. hispidula (Less.) Beauverd, A. perfoliata and A. semipapposa (DC.) Beauverd. The closest affinities of A. arctotoides and the closely related A. hispidula are indicated to be with Cymbonotus. Lewin (1922a)
placed A. semipapposa in A. sect. Acuminatae K.Lewin.
As Arctotis and Cymbonotus are presently defined, cypselar morphology and filament ornamentation are insufficient to support the placement of species of Cymbonotus and Arctotis sect. Austro-orientales in separate genera. Our results do not support the merging of Cymbonotus and Arctotis by Beauverd (1915)
, but rather support expanding the circumscription of Cymbonotus by transferring species currently placed in Arctotis sect. Austro-orientales as well as Haplocarpha schimperi. However, a more detailed molecular study including a broader sample of taxa is required before taxonomic changes can be contemplated and an ongoing investigation of floral morphology might help to clarify relationships further.
Dymondia
Recognition of the monotypic genus Dymondia is strongly supported by both the cpDNA and ITS data sets. The Dymondia margaretae lineage is indicated to have diverged prior to the main phase of diversification that gave rise to the Arctotheca, ‘Arctotis', ‘Cymbonotus', and ‘core Haplocarpha' lineages. Dymondia margaretae is distinctive in Arctotidinae in having sessile capitula, disc florets with lobed stigmatic arms, and often retuse anther appendages. Its rhizomatous habit is shared with Haplocarpha scaposa. Mature cypselae have never been seen, but the ovaries are glabrous and have a well-developed basal coma of twin hairs, and a biseriate pappus of free scales is present.
Haplocarpha
The cpDNA and ITS data sets provide strong evidence that Haplocarpha, as it is presently circumscribed, is polyphyletic. The six Haplocarpha species included in this study were split between four well-supported clades. Haplocarpha nervosa and H. rueppellii comprised the earliest diverging lineage in all phylogenies with maximum support. Haplocarpha scaposa was placed as a maximally supported monotypic lineage also diverging early in diversification of the Arctotidinae. Haplocarpha lanata and H. lyrata formed a terminal clade, also including Arctotheca calendula in the cpDNA phylogeny. Surprisingly, Haplocarpha schimperi was nested within a well-supported clade also containing Arctotis arctotoides and Cymbonotus lawsonianus.
The type species of Haplocarpha is H. lanata. The phylogenies presented herein indicate H. lanata and H. lyrata are phylogenetically close and belong to a ‘core Haplocarpha' clade that might also include Arctotheca calendula. Morphological evidence (Lewin, 1922a
; McKenzie et al., 2005
) indicates the unsampled H. oocephala (DC.) Beyers and H. parvifolia, which have restricted distributions in the Western Cape province to the north of H. lanata, might also belong in the ‘core Haplocarpha' group. This group is characterized by possessing a tufted or prostrate growth habit, female ray florets and hermaphrodite disc florets, erect nude peduncles exceeding the leaves in length when the cypselae are mature, cypselae with ribs more prominent on the abaxial surface than the adaxial surface, a well-developed pappus of free scales in one whorl and exceeding the cypsela in length, and a basal coma of twin hairs exceeding the cypsela in length. Haplocarpha oocephala and H. parvifolia differ from H. lanata and H. lyrata in possessing cypselae with very broad abaxial ridges and lacking twin hairs on the cypsela surface (McKenzie et al., 2005
). In addition, H. parvifolia has smooth, rather than papillose, filaments (Lewin, 1922a
).
The molecular data presented herein provide strong evidence that H. nervosa and H. rueppellii do not belong in the ‘core Haplocarpha' group. Cypselar morphology also supports this conclusion (McKenzie et al., 2005
). Previously, Haplocarpha nervosa and H. rueppellii have been segregated in the genus Landtia (Lessing, 1831
, 1832
; Harvey, 1865
; Bentham, 1873a
, b
). Some morphological evidence (Lewin, 1922a
; McKenzie et al., 2005
) indicates the unsampled H. hastata K. Lewin might also belong in the Landtia group. A variety of morphological characters have been proposed to support recognition of Landtia, including peduncle length relative to leaf length, cypsela shape and surface sculpturing, pappus scale apex shape, and pappus length relative to cypsela length (e.g., Harvey, 1865
; Bentham, 1873a
, b
; Phillips, 1951
). Other authors have concluded the characters utilized are untenable to support recognition of Landtia (Beauverd, 1915
; Lewin, 1922a
). Therefore, in more recent treatments Landtia is subsumed within Haplocarpha (Norlindh, 1977
; Bremer, 1994
). However, since Bentham's (1873a
, b
) treatment, H. schimperi has been placed in the Landtia–Haplocarpha group. The sequence data presented herein, in conjunction with cypselar morphology (McKenzie et al., 2005
), indicate that H. schimperi is not closely related to either ‘core Haplocarpha' or Landtia. The exclusion of H. schimperi permits ‘core Haplocarpha' and Landtia to be discriminated on morphology more readily.
Haplocarpha nervosa and H. rueppellii are prostrate or rosulate herbaceous perennials usually inhabiting damp situations such as boggy areas and streambanks. They possess geocarpic capitula (Barker, 2005
), smooth filaments, cypselae with a convex adaxial surface and abaxial ridges barely more conspicuous than the adaxial ridges, a very reduced pappus of free scales not conspicuously differentiated into distinct series, and the basal coma of twin hairs is very reduced or absent. Haplocarpha hastata is anomalous in that the cypselae are glabrous, bilaterally flattened and obovoid, the pappus scales are fused, and a coma is absent. In addition, the leaf morphology of H. hastata is very different. It is unknown whether the capitula of H. hastata exhibit geocarpy. Haplocarpha hastata is distinguished from ‘core Haplocarpha' in possessing smooth filaments, cypselae lacking conspicuous abaxial ridges and a coma, and a very reduced pappus of fused scales.
Given the results of the present study, the weak development of ridges on the abaxial surface of the cypsela in the Landtia group (McKenzie et al., 2005
) and the distinct chromosome number of H. rueppellii (2n = 30) compared to all other reported chromosome counts in the Arctotidinae (including counts for H. lyrata and H. scaposa) of 2n = 18 (e.g., Norlindh, 1965
, 1977
; Strother et al., 1996
; Carr et al., 1999
; Skinner, 2001
; Mitchell, 2005
) are interpreted as supporting the basally diverging position of the Landtia lineage in Arctotidinae. The very reduced pappus and coma on the cypsela might be synapomorphic and related to the evolution of geocarpic capitula in the lineage.
Haplocarpha scaposa is indicated to represent a monotypic lineage diverging relatively early in the diversification of Arctotidinae. Cypsela anatomy is consistent with the molecular data in indicating H. scaposa has a closer affinity with H. rueppellii than with ‘core Haplocarpha' and other genera of Arctotidinae (Reese, 1989
). However, external cypselar morphology (McKenzie et al., 2005
) and the possession of shortly papillose filaments indicates an affinity with ‘core Haplocarpha'. Haplocarpha scaposa is unusual in possessing a rhizomatous growth habit, a trait shared with D. margaretae.
In both the cpDNA and ITS phylogenies, H. schimperi is nested in a clade with A. arctotoides and C. lawsonianus. Such a relationship has not been suggested previously, although the generic affinities of H. schimperi have long been uncertain. Originally placed in Schnittspahnia Sch.Bip. with H. rueppellii by Schultz Bipontinus (1842)
, H. schimperi subsequently has been placed in the monotypic genus Ubiaea J. Gay (Gay, 1847
), Landtia (Bentham, 1873a
, b
) and Haplocarpha subg. Landtia (Beauverd, 1915
). Certain morphological evidence conflicts with the molecular phylogenetic hypotheses. The cypselae of H. schimperi are always pappose, but no member of A. sect. Austro-orientales or Cymbonotus has even rudimentary pappus scales (McKenzie et al., 2005
). A pappose cypsela might be plesiomorphic in this group. Alternatively, the coronate pappus in some H. schimperi plants is shared with the fellow Ethiopian H. hastata and so could be indicative of introgression. The capitula of H. schimperi are geocarpic (Barker, 2005
). The absence of abaxial cypselar wings in H. schimperi might represent a secondary reduction associated with geocarpy. The cypselae of H. schimperi show little morphological similarity to those of ‘core Haplocarpha' and differ from those of the Landtia group in being bilaterally flattened, possessing well-developed, dentate abaxial ridges and a conspicuously rugose surface. The cypselae superficially are similar to those of a form of Arctotis fastuosa and Cymbonotus preissianus, but H. schimperi differs from these species in possessing a pappus and the abaxial ridges are not developed into wings (McKenzie et al., 2005
). An ongoing study of floral morphology in the Arctotidinae might further clarify the affinities of H. schimperi.
Biogeographic implications
The biogeography of Arctotidinae will be discussed more fully in a forthcoming paper. Consequently, only a brief summary of the biogeographic implications of the phylogenetic hypotheses will be noted here.
The Cape Floristic Region (CFR; sensu Goldblatt and Manning, 2002
) in southwestern South Africa possesses a distinctive flora indicated to be derived from lineages presently occurring on most continents, therefore the geographic sources of CFR clades are likely to be varied (Galley and Linder, 2006
). Although the main center of diversity of Arctotidinae is in the winter-rainfall CFR, the sampled species in the most basal extant lineages (the ‘Landtia clade' and Haplocarpha scaposa) occur outside this area in the summer-rainfall region of southern and eastern Africa (Fig. 2). The ‘Landtia clade' (represented by H. nervosa and H. rueppellii) forms a chain of disjunct populations on the southern and eastern African mountains. Haplocarpha nervosa occurs in South Africa from George in Western Cape Province to the Drakensberg Mountains, with an outlying population on the Nyanga Mountains, Zimbabwe (Hilliard, 1971
, 1977
; Pope, 1992
; Goldblatt and Manning, 2000
). Haplocarpha rueppellii occurs on high-altitude mountains from Tanzania to the Ethiopian highlands (Hedberg, 1957
). This distribution area coincides with White's (1983)
Archipelago-like Afromontane-Afroalpine Floristic Region, which comprises isolated temperate "islands" above the warmer and drier surrounding plains and provides a high-altitude corridor linking South Africa and tropical eastern Africa. Haplocarpha scaposa has the widest distribution of any species in Arctotidinae, occurring in mesic and submontane grassland in eastern South Africa and the Zambezi River catchment area in tropical eastern Africa (Hilliard, 1977
; Pope, 1992
).
Goldblatt (1978)
listed 19 angiosperm genera that, like Arctotidinae, have a southern African–East African mountain chain distribution centered in the CFR and decreasing northwards. The ancestral areas of such genera are still unknown, but could constitute either the CFR, temperate southern Africa, tropical Africa, or other continents. In an analysis of CFR clades for which unambiguous geographic area relationships could be determined, Galley and Linder (2006)
found that almost half the clades tested exhibited trans-Indian Ocean relationships and in only two cases (Penaeaceae and Stilbaceae) a sister relationship between tropical Africa and the CFR was established. In eight CFR clades containing afromontane species, Galley and Linder (2006)
found that the afromontane species were nested within the clade and thus dispersal from the CFR is indicated. In contrast, in Arctotidinae the two basal lineages comprised afromontane–tropical eastern African species and only one eastern African species, H. schimperi, is nested within the subtribal clade. One interpretation is that Arctotidinae might have originated in a summer-rainfall regime in temperate southern or eastern Africa. Alternatively, ancestral Arctotidinae might have been widespread in southern Africa and have diversified vicariously. However, we emphasize that more comprehensive sampling and distribution data are needed before hypotheses regarding the ancestral area and subsequent dispersal and radiation of Arctotidinae can be tested.
Species from the CFR and Australia are represented as derived within Arctotidinae, and two reinvasions of the Eastern Cape and KwaZulu-Natal provinces (by Arctotis arctotoides and Haplocarpha lyrata) are indicated. The sampled species in the ‘Arctotis' clade occur in the southern African winter-rainfall region except for A. venusta, which is distributed in the semi-arid and summer-rainfall regions of Namibia, southern Botswana, and northern South Africa (Norlindh, 1965
; Pope, 1992
). Arctotheca calendula, Arctotis breviscapa, A. sect. Anomalae, and Haplocarpha lanata grow in the winter-rainfall region of South Africa. Both Arctotis perfoliata and H. lyrata occur in the transition zone between the winter- and summer-rainfall regions, but A. perfoliata also extends westwards to the Cape Peninsula. Dymondia margaretae, which this study indicates is an early divergence among the CFR-endemic taxa, has a localized distribution on the Agulhas Plains (Compton, 1953
; Rourke, 1974
).
Major geographic disjunctions exist within the ‘Cymbonotus' clade. Cymbonotus lawsonianus is endemic to eastern and southern Australia and Tasmania (Murray, 1992
). Haplocarpha schimperi is distributed from Eritrea to Kenya, occurring at moderate to high altitudes. Arctotis arctotoides belongs to a southern African species complex centered in the Eastern Cape province, Highveld and the Drakensberg (S. D. Mitchell, R. J. McKenzie, and N. P. Barker, Rhodes University, unpublished data). This complex corresponds to Lewin's (1922a) Arctotis sect. Austro-orientales but with A. perfoliata (and its synonym A. discolor) excluded. Two species within this complex, A. microcephala (DC.) Beauverd and A. scapiformis Thell., occur at high altitudes in the Eastern Cape and Drakensberg Mountains and thus
