|Home | About | Journals | Submit | Contact Us | Français|
Section Acrocentron of the genus Centaurea is one of the largest sections of Centaurea with approx. 100 species. The geographic distribution, centred in the Mediterranean, makes it an excellent example for studies of the biogeographic history of this biodiversity-rich region.
Plastid (trnH-psbA) and nuclear (ITS and ETS) DNA sequence analysis was used for phylogenetic reconstruction. Ancestral biogeographic patterns were inferred by dispersal-vicariance analysis (DIVA).
The resulting phylogeny has implications for the sectional classification of Acrocentron and confirms merging sect. Chamaecyanus into Acrocentron as a subsection. Previous suggestions of an eastern Mediterranean origin of the group are confirmed. The main centres of diversification established in previous studies are now strongly supported. Expansion of the group in two different radiations that followed patently diverse paths is inferred.
Radiation followed two waves, widely separated in time scale. The oldest one, from Turkey to Greece and the northern Balkans and then to North Africa and Iberia, should be dated at the end of the Miocene in the Messinian period. It reached the Iberian Peninsula from the south, following a route that is landmarked by several relictic taxa in Sicily and North Africa. A later radiation during the Holocene interglacial periods followed, involving species from the north of the Balkan Peninsula, along a Eurasian pathway running from Central Iberia to the steppes of Kazakhstan. A generalized pattern of reticulation is also evident from the results, indicating past contacts between presently separated species. Molecular data also confirmed the extent of hybridization within Acrocentron and were successful in reconstructing the paleogeography of the section.
The tribe Cardueae (Compositae) has been deeply reviewed and widely modified in the past 10 years. Most of the many systematic problems within the tribe, from the subtribal classification to the delineation of genera, are now more clear (Susanna et al., 1995, 2006; Garcia-Jacas et al., 2001). This has been a complicated task because the tribe includes some of the largest genera of the whole family (Centaurea L., Cirsium Mill., Cousinia Cass. and Saussurea DC.). Our present research centres in one of these large genera, Centaurea.
In its present delimitation as adopted in the latest revision of the Cardueae (Susanna and Garcia-Jacas, 2007), Centaurea comprises approx. 250 species (from 400 in earlier classifications; e.g. Dittrich, 1977). Centaurea encompasses only the large Jacea group (revised by Garcia-Jacas et al., 2006), the section Cyanus and the Acrocentron group.
The Acrocentron group is formed by Centaurea sect. Acrocentron (Cass.) DC. with the addition of two much smaller sections: Centaurea sect. Chamaecyanus Willk. and C. sect. Stephanochilus Coss. & Dur. A previous proposal by Wagenitz and Hellwig (1996) also included Centaurea sect. Crocodylium DC. and Centaurea sect. Aegialophila (Boiss. & Heldr.) O. Hoffm., but these two sections should be excluded and merged in a different genus, Crocodylium Cass. (Garcia-Jacas et al., 2001; Vilatersana et al., 2001; Font et al., 2002). Among other characters, the Acrocentron group is well deﬁned by having one of the four pollen types described in the genus Centaurea, the Centaurea scabiosa pollen type (Wagenitz, 1955). A previous molecular survey of the group using sequences of the ITS region (Font et al., 2002) only confirmed the inclusion of sect. Stephanochilus and the exclusion of the restored genus Crocodylium, and suggested merging C. sect. Chamaecyanus in C. sect. Acrocentron. However, for the rest, the resulting trees consisted largely of unsupported polytomies. The largest part of the Acrocentron group is Centaurea sect. Acrocentron, a huge section with approx. 100 species mainly in the Mediterranean region.
An aspect of special interest to the present contribution is the history of expansion of Centaurea sect. Acrocentron through the Mediterranean. Reconstruction of the paleogeographic history of plants is a relatively novel focus in botany, and the present distribution of C. sect. Acrocentron in the Mediterranean region makes the group an excellent object of study. In our previous molecular survey, the existence of three main centres of speciation (Anatolian–Aegean, North Balkanic–Eurasian, and Iberian–North African) was suggested. However, as previously noted, the resulting phylogeny was flawed by an extremely low support for most of the clades. A detailed phylogeny of the related Centaurea group Jacea (Garcia-Jacas et al., 2006) confirmed the existence of the three centres of speciation hinted at in Font et al. (2002). A recent paper dealing with the related Acrolophus–Phalolepis complex of Centaurea has contributed to our knowledge of the expansion of Centaurea in the Iberian Peninsula and North Africa (Suárez-Santiago et al., 2007). It is tempting to verify whether Acrocentron follows similar patterns of colonization of the Mediterranean.
The general lack of support in our previous report was considered a result of intense historical gene flow (Font et al., 2002). Hybridization in the Acrocentron group is well documented (Kummer, 1977; Fernández Casas and Susanna, 1986; Garcia-Jacas and Susanna, 1993, 1994; Garcia-Jacas, 1998; Hellwig et al., 1994). Hybrids are often homoploid (Kummer, 1977; Fernández Casas and Susanna, 1986), which favours gene flow because homoploid hybrids act as a bridge between the parental species (McKinnon, 2005).
Based on these previous results we decided to carry out a new study with the addition of the ETS region of the nuclear ribosomal DNA and the non-coding intergenic spacer trnH-psbA of plastid DNA. Incongruence between nuclear and plastid phylogenies is an important indicator of historical interspecific gene flow (Wendel and Doyle, 1998). The ETS region has proved to be a good tool for the phylogenetic reconstruction of closely related species in many genera from the Compositae (Calycadenia, Baldwin and Markos, 1998; Helianthus, Linder et al., 2000; Lessingia, Markos and Baldwin, 2001; Xylothamia and Gundlachia, Urbatsch et al., 2003; Montanoa, Plovanich and Panero, 2004) and has already been very useful in the Cardueae–Carduinae (Kelch and Baldwin, 2003) and Cardueae–Centaureinae (Hidalgo et al., 2006). Moreover, the ETS region has demonstrated great potential for unraveling the extent of gene flow in Centaurea sect. Willkommia (Suárez-Santiago et al., 2007). Our goals with this combined approach are as follows: (1) to examine whether a phylogenetic subsectional classification of Centaurea sect. Acrocentron is supported by molecular data and to confirm the inclusion of C. sect. Chamaecyanus in sect. Acrocentron with subsectional rank; (2) to seek for molecular evidence of hybridization and gene flow among C. sect. Acrocentron; (3) to verify the existence of the suggested centres of speciation of the group; and (4) to explore and reconstruct the history of the expansion of C. sect. Acrocentron through the Mediterranean and establish the centre of origin of the group.
The ETS (external transcribed spacer) and ITS (internal transcribed spacer) regions of the nrDNA of 45 taxa of Centaurea sect. Acrocentron and C. sect. Chamaecyanus were sequenced. In some cases, more than one population was sequenced. Our ITS1 and ITS2 sequences from previous work (Font et al., 2002) were complemented by sequencing the 5·8S region. Sampling included all the geographical areas of distribution of the section: eastern Mediterranean (Balkan Peninsula and Anatolia), Irano-Turanian, western Mediterranean (Iberian and Italian Peninsulas, and North Africa) and some species of wide distribution, such as Centaurea scabiosa L.
A plastid non-coding region, the intergenic spacer trnH-psbA, was also sequenced. In view of the very low number of informative characters found in this region, only a representative subset of 31 taxa was sequenced (Table 1).
Outgroups were chosen among the genera of the ‘derived clade’ of the subtribe Centaureinae (Garcia-Jacas et al., 2001), where the Acrocentron group belongs, and included one species each from the Cyanus and Jacea groups of Centaurea and the genera Carduncellus Adans. and Phonus Hill.
Voucher data, source and GenBank sequence accession numbers are given in Table 1.
Total genomic DNA was extracted following the CTAB method of Doyle and Doyle (1987) and Cullings (1992) from silica-gel-dried leaves collected in the field. In some cases, herbarium material was used.
Double-stranded DNA of the ITS region was amplified using ITS1 (White et al., 1990) and 1406F (Nickrent et al., 1994) as the forward primer and ITS4 as the reverse primer (White et al., 1990). The profile used for PCR amplification follows the protocol described in Susanna et al. (2006). The ETS region was amplified with ETS1F as the forward primer and as reverse primer either 18S-2L (Linder et al., 2000) or 18SETS (Baldwin and Markos, 1998). The PCR was executed with the following conditions: 5 min denaturing at 95 °C, followed by 30 cycles of 94 °C denaturing for 45 s, 48 °C annealing for 45 s and 72 °C extension for 40 s, with an additional 7 min at 72 °C. PCR products of the species with more than one band were cloned using a TOPO TA Cloning Kit (Invitrogen, Carlsbad, CA) following the manufacturer's instructions, except that only half of the reaction volume was used. When possible, from eight-to-fifteen positive colonies from each reaction were screened with direct PCR using T7 and M13R universal primers following the protocol of Vilatersana et al. (2007). Five-to-eight PCR products were selected for sequencing in both directions using the same primers.
Plastid and nuclear PCR products were purified with the QIAquick PCR Purification Kit (Qiagen Inc., Valencia, CA). Direct sequencing of the amplified DNA segments was performed using a BigDye Terminator Cycle Sequencing v3.1 (Applied Biosystems, Foster City, CA), following the manufacturer's protocol. Nucleotide sequencing was performed on an ABI PRISM 3100 DNA Analyzer (Applied Biosystems). When cloned sequences from one accession presented only a single substitution with regard to the others, they were excluded. Consensus sequences were generated for each accession and region, condensing the single base-pair differences among cloned sequences into a degenerated base. This strategy reduced the size of the matrices as well as the impact of potential PCR artifacts (chimerical sequences and Taq errors; Cline et al., 1996; Popp and Oxelman, 2001).
Sequences were aligned visually by sequential pairwise comparison (Swofford and Olsen, 1990). The data matrices are available on request from the corresponding author. In view of the practical impossibility of unambiguously aligning the most variable 5′ end of the ETS region of the ingroup and the outgroup, two analyses were carried out. First, we analysed the more conserved 3′ETS region plus the ITS region including the outgroups, with the aim of identifying the earliest-branching clade of Acrocentron. Second, we carried out an analysis of the complete-region ETS plus ITS, including only the species of the Acrocentron group, using as outgroup the clade identified as earliest-branching in the first analysis. For the analysis of the trnH-psbA plastid region, we used as outgroup some representatives from the clade identified as earliest-branching in the first analysis.
Phylogenetic analyses were conducted using both parsimony and Bayesian inference optimality criteria. The plastid dataset was analysed independently. Both nrDNA regions (ETS plus ITS) were combined and analysed in one matrix because both datasets resulted in fully compatible topologies. A partition homogenity test was impossible to complete because the heuristic search collapsed after reaching the tree limits of PAUP in the first replicate.
Parsimony analysis involved heuristic searches conducted with PAUP ver. 4.0b10 (Swofford, 2002) using tree-bisection-reconnection (TBR) and branch swapping with character states specified as unordered and unweighted. The indels were treated as missing data in all the analyses. All most-parsimonious trees were saved. To locate other potential islands of most-parsimonious trees (Maddison, 1991) 100 replications were performed with random taxon addition, also with TBR branch swapping. Consistency index (CI) and retention index (RI) are always given excluding uninformative characters. Bootstrap analysis (BS) was performed in order to assess the confidence of the branches (Felsenstein, 1985). For the combined matrices (ETS plus ITS), bootstraping followed the approach of Lidén et al. (1997) using 1000 replicates, random taxon addition with ten replicates, and no branch swapping.
For the Bayesian analysis, the data sets were analysed using MrModeltest ver. 2.2 (Nylander, 2004) to determine the sequence evolution model that best described the data. This model was used to perform a Bayesian analysis using the program MrBayes ver. 3.1.2 (Huelsenbeck and Ronquist, 2001; Ronquist and Huelsenbeck, 2003). Four Markov chains were run simultaneously for 1 000 000 generations and sampled every 100 generations. Data from the first 1000 generations were discarded as the ‘burn-in’ period, after confirming that likelihood values had stabilized prior to the 1000th generation. The 50 % majority rule consensus phylogeny and posterior probability of nodes (PP) were calculated from the remaining sample.
Ancestral biogeographic reconstruction was inferred with the dispersal-vicariance analysis implemented in DIVA (Ronquist, 1996, 1997). Eight areas of endemism were defined by the presence of one or more endemic taxons: A = Iberian; B = North African; C = Balkanic; D = Aegean; E = Anatolian + Iranian; F = Eurasian; G = Caucasian; and H = Sicilian. Because DIVA needs a dichotomic tree, we used as input data the tree with the highest posterior probability, obtained with the first analysis (3'ETS plus ITS data; see Fig. 1). Outgroup taxa were omitted when running the DIVA because these taxa were not biogeographically informative. The analysis was carried out constraining the number of ancestral areas to a maximum of two, which is the maximum geographic range of the most widespread species.
Numerical results of the analyses are shown in Table 2. The analysis of the 3'ETS plus ITS regions is shown in Fig. 1. Centaurea sect. Acrocentron group is monophyletic with high support (BS = 100 %, PP = 1·0) only if we include C. sect. Chamaecyanus, which is not confirmed as a natural group. The earliest-branching group of Acrocentron is the clade of the Anatolian and Irano-Turanian taxa (clade ANAT), without parsimony support but with very high Bayesian suport (BS = 58 %, PP = 1·0). Next, there comes the Aegean taxa (clade AEG) almost without support (BS = 58 %, PP = 0·95). Some species of wide distribution (the Centaurea scabiosa group) ranging from the Iberian Peninsula to Central Asia, together with a group of taxa from the north Balkans and north Italy (clade Balkans), are sister to the rest (BS = 65 %, PP = 1·0). Finally, there is a dichotomy comprising two clades. The first subclade is a well-supported group (BS = 80 %, PP = 1·0) with all the species from North Africa and Sicily and some others from the Iberian Peninsula (clade North Africa). The second subclade (BS = 96 %, PP = 1·0) includes only species from the Iberian Peninsula.
The analysis of the complete ETS plus ITS region (Fig. 2) shows an important difference with the partial analysis: the Iberian Peninsula and North African clades, which formed a well-supported clade in the first analysis (Fig. 1), are not grouped together in the complete analysis. In this second analysis, the Balkans–Eurasia clade is sister to the Iberian Peninsula clade, and this joint clade is sister to the North African clade (Fig. 2). Besides this divergent result, a few additional differences become apparent. The cloned sequences of C. josiae Humbert form a monophyletic clade in the partial analysis (Fig. 1), but one of the cloned sequences is associated with C. amourensis Pomel in the complete analysis (Fig. 2). One of the cloned sequence of C. argecillensis Gredilla (species from Central Iberia), which in the partial analyses occupies an unresolved position at the base of the North African clade (Fig. 1), is grouped together with C. litardierei Jahand. & Maire in the complete analysis (Fig. 2). There is also a difference in the group formed by the cloned sequences of Portuguese C. prolongi Boiss. (Pop. 3), which appear on different clades in the partial analysis (clade Iberian Peninsula, Fig. 1) and unsupportedly united in the complete analysis (clade Iberian Peninsula, Fig. 2). Finally, C. lainzii Fern. Casas, which appears as an isolated taxon in the partial analysis (clade North Africa, Fig. 1), is united with C. prolongi (Pop. 1 and 2) and C. haenseleri Boiss. in the complete analysis (clade North Africa, Fig. 2). Differences in the topology resulting from the two analyses (ITS + ETS 3′, Fig. 1; ITS + ETS 5′ and 3′, Fig. 2) are caused by the number of informative characters yielded by the markers used in each case – much higher for the ETS 5′ end.
The complete analysis (Fig. 2) suggests that the species from the Aegean are closer to the outgroup formed by the Anatolian and Irano-Turanian species. Next, there come two clades, one with species from the Balkans and some others of wide distribution (clade Balkans–Eurasia), and the taxa from the Iberian Peninsula and a clade with species of North Africa, Sicily and the rest of the Iberian taxa (clade North Africa) with high support both from BS and PP.
The low number of informative characters in the plastid dataset (Table 2) explains the lack of resolution in the phylogenetic analysis (tree not shown). Table 3 shows the changes of the sequences in the plastid dataset.
The output of the DIVA is the most-parsimonious ancestral area for each node, and the results are reflected in Fig. 1. Three equally parsimonious solutions for the ancestral area are suggested for the deepest node (Iberian/Anatolian, Balkanic/Anatolian and Aegean/Anatolian). This ancestor would have then dispersed to the Iberian/Aegean or Balkanic/Aegean areas. The ancestral area suggested for the third deepest node is Iberian/Balkanic. Beyond this node, the DIVA results suggest predominance of vicariance.
The presence of different rybotypes in both regions (ETS and ITS) confirms that in some species of the group concerted evolution has not occurred or has been incomplete. This is an acute contrast with the related Centaurea sect. Willkommia in which biased concerted evolution was extensive (Suárez-Santiago et al., 2007).
The classic taxonomic classification in two sections, Acrocentron and Chamaecyanus, is untenable after our results. The segregation of Centaurea sect. Chamaecyanus (species of this section are marked CH in Figs 1 and and2)2) would leave C. sect. Acrocentron paraphyletic. The most appropriate status for C. sect. Chamaecyanus is subsectional within C. sect. Acrocentron, confirming previous suggestions (Font et al., 2002). In the Iberian clade (Figs 1 and and2),2), species of both sections are part of a polytomy, and C. sect. Chamaecyanus is not resolved as a natural group. However, both sections are morphologically very well defined (Fernández Casas and Susanna, 1986), and we suspect that the lack of confirmation of the monophyly of C. sect. Chamaecyanus is caused by introgression. Intermediates between them are very frequent (Fernández Casas and Susanna, 1986), a fact that is probably the cause of this entanglement and the reason for the low support for the clade that comprises many of the species of C. sect. Chamaecyanus (Figs 1 and and2).2). The position of Centaurea crocata Franco as sister to the Iberian clade is in this sense very significant: on the basis of morphology, C. crocata is suspected as a hybrid between taxa of both sections (Garcia-Jacas and Susanna, 1994), which would explain this basal position.
Regarding the possibility of establishing a subsectional classification within C. sect. Acrocentron as suggested by previous workers (Hayek, 1901; Routsi and Georgiadis, 1994), molecular data are of little help because intense gene flow hinders molecular reconstruction of phylogenetic affinities, and the extreme homogeneity of morphological characters makes any suggestion of infrasectional classification a very difficult task. As is evident from Figs 1 and and2,2, molecular data group the taxa on a geographical basis.
In our previous study based on the ITS region (Font et al., 2002), we suggested that the very low definition within Centaurea sect. Acrocentron was probably caused by the high levels of hybridization and reticulation within the group. Our new results could support this view, for example the presence of an ETS cloned sequence of C. argecillensis from central and north-east Iberia within the North African clade, whilst the rest of the cloned sequences of this species are placed in the Iberian clade (Figs 1 and and2).2). There are some other examples, perhaps not so extreme but also indicating intense gene flow: Centaurea toletana Boiss. & Reut. (2n = 4x) has a cloned sequence united to C. podospermifolia Loscos & J. Pardo with high PP support, and another one ranged within the Iberian polytomy (Figs 1 and and2).2). Another case is a population of C. prolongi from Portugal, firstly cited as C. collina L., later on identified as C. prolongi (Garcia-Jacas and Susanna, 1994) and even considered a new species, C. occasus Fern. Casas (Fernández Casas, 1997). The two cloned sequences of this population of C. prolongi (Pop. 3) appear in the same clade, but in different positions (C. prolongi 3, clade Iberian Peninsula; Fig. 1).
Gene flow could also explain the differences within C. prolongi: the Portuguese population is associated with the taxa from central and north Iberia, whilst populations from the extreme south of Spain are placed in the African clade (C. prolongi Pop. 1 and Pop. 2, clade North Africa; Figs 1 and and2).2). This segregation shows that they have different ribotypes, a fact that could reinforce the hypothesis that they are different species. However, morphological differences alleged by Fernández Casas (1997) are only quantitative (namely, the size of the leaves and florets), which could easily be explained by the different climatic conditions of both localities (milder and wetter in Portugal). A prefereable explanation for the positions of populations of C. prolongi in opposite clades is gene flow: populations of C. prolongi in south Spain are very close to the African coast (Fig. 3), and other species of C. sect. Acrocentron growing nearby (C. haenseleri and C. lainzii) also appear in the African group (Figs 1 and and2).2). In contrast, the studied Portuguese locality was far apart from the waves of African taxa.
In our analyses, C. legionis-septimae Fern. Casas & Susanna, a narrow endemic from the mountains of northern Spain, is united with C. cephalariifolia Willk., a species from the C. scabiosa widespread group. This position contradicts morphologic evidence, which relates C. legionis-septimae to C. ornata Willd. (Garcia-Jacas and Susanna, 1992), and could be considered another indication of hybridization, as suggested by the existence of homoploid hybrid populations between both taxa described as C. ×cephalariiseptimae Fern. Casas & Susanna (Fernández Casas and Susanna, 1985).
The position of the Balkan clade, which is united to the Iberian clade in the phylogenetic analysis (Fig. 2), could be explained in terms of hybridization and introgression. This relationship could be caused by gene flow of the widely distributed taxa from the clade Balkans–Eurasia and the Iberian species. The culprit is certainly C. cephalariifolia, a tetraploid race of the widespread diploid C. scabiosa. To date, up to six naturally occurring hybrids have been reported between C. cephalariifolia and other Iberian taxa (we have discussed one of them above, C. ×cephalariiseptimae). In four cases, the crosses involve tetraploid parentals and the hybrids are also tetraploid (Kummer, 1977; Fernández Casas and Susanna, 1986) – thereafter they are not genetically isolated from their progenitors.
A final reason for considering high levels of introgression is the result of the plastid sequencing (Table 3). The few changes observed in the sequences of the intergenic spacer trnH-psbA do not agree with any known relationship, either taxonomic or biogeographic. Cytoplasmic gene flow could be one of the main causes of low resolution in phylogenies (Rieseberg and Soltis, 1991) or, alternately, C. sect. Acrocentron shows high levels of homoplasy in the trnH-psbA (Hamilton et al., 2003).
The combined ITS plus ETS phylogeny (Figs 1 and and2)2) shows that there are five speciation centres: Anatolia–Iran, the Aegean (these two were united into one by Font et al., 2002), the Balkans (including some species of wide Eurasian distribution), the Iberian Peninsula, and North Africa. These geographic denominations for the clades are general and do not imply that all the taxa within the clade grow in this precise geographic area, but they are accurate enough for the purpose of briefly discussing each centre and the history of the geographic distribution of C. sect. Acrocentron (Fig. 3).
Distribution of Acrocentron has been presented as an east-to-west migration (Garcia-Jacas and Susanna, 1992). As a confirmation of this hypothesis, the Anatolian–Iranian centre is the earliest-branching clade in the first of our analyses, 3'ETS plus ITS (Fig. 1). This is one of the three hypotheses suggested by the DIVA, and indeed the most likely because the two areas are contiguous. The alternate solutions suggest either an unlikely Iberian–Anatolian origin for the group or a Balkanic–Anatolian origin that would obviate the Aegean milestones. The origin of Acrocentron should be placed in the eastern Mediterranean, much like the rest of the subtribe according to the greater richness of genera of the Centaureinae in this area (Susanna and Garcia-Jacas, 2007). The hypothesis of the south of the Irano-Turanian region (Iranian plateau) as the place of origin of the group was rejected on karyologic grounds, and Iranian species of C. sect. Acrocentron were considered a later expansion from Anatolian populations (Garcia-Jacas et al., 1998).
From Anatolia, species of C. sect. Acrocentron radiated to the Aegean islands, the species of which (C. aetholica Phitos & T. Georgiadis, C. ebenoides Heldr., C. euboica Rech. f. and C. raphanina Sibth. & Sm.) form a well-supported clade (AEG in Fig. 1 and Aegean in Fig. 2) that is sister to the next branches. This radiation is supported by the DIVA results, which suggests two equally possible solutions for this dispersal: an unlikely (not contiguous) Iberian–Aegean ancestral area, or the Balkanic–Aegean hypothesis (Fig. 1). The Aegean islands and south Greece constitute the easiest way of penetration from Anatolia to the western Mediterranean, which is consistent with the intermediate position of the Aegean clade in the phylogeny (Figs 1 and and22).
The Aegean clade is at the basis of the two main radiations of C. sect. Acrocentron, as reflected in our phylogeny (Fig. 1). In the north of its range, it originated the Balkanic clade, and in the south it migrated to the extreme south of the Italian Peninsula (Fig. 3). A dual radiation is supported by the DIVA results (Fig. 1), which suggest an Iberian–Balkanic ancestral area for this node. Predominance of Iberian species in the upper part of the tree could be the reason for DIVA assigning an Iberian origin to this branch, which should be better considered as North African–Balkanic.
The Messinian salinity crisis (between 6 and 5·2 MYA) offers a good opportunity for hypothesizing a date for the migration of the Acrocentron group. In the east, a pass between Anatolia and the Aegean was already open by the middle Miocene (Çağatay et al., 2006). The lowering of the sea level opened the way from the Aegean and Greece to the extreme south of Italy and Sicily, and then to North Africa and the Iberian Peninsula (Fig. 3). The main migration of C. sect. Acrocentron through this southern pathway probably would have terminated in the Pliocene because species of C. sect. Acrocentron are absent in all the western Mediterranean archipelagos, which during the Messinian were connected to the mainland. This dating has also been suggested for the expansion of the related complex Acrolophus–Phalolepis of the genus Centaurea (Suárez-Santiago et al., 2007), a group highly speciated in Iberia but absent from the Balearic Islands.
Besides paleogeographic arguments, a series of landmarks illustrate this hypothetical path. Four species closely related on morphological grounds mark the southern path of the migration: C. tauromenitana Guss. from Sicily, C. carolipauana Fern. Casas & Susanna from northern Morocco, C. clementei Boiss. from northern Morocco and south Iberia and C. lainzii from south Iberia (Fig. 3). All of them are relictic mountain plants and, with the exception of C. clementei, are known from a single locality. They are chasmophytes sharing some unusual characters in C. sect. Acrocentron, such as very wide involucral bracts ending in an unarmed black appendage (Fernández Casas and Fernández Morales, 1979). Figure 3 summarizes the hypothesized pathes of expansion of the group, together with the distribution of these landmark species. The expansion of the group to the rest of the Iberian Peninsula and North Africa was a natural consequence, favoured by the development of a Mediterranean climate. The poverty in species of C. sect. Acrocentron in the North African lowlands stretching between Algeria and Libya reflects the deep changes in the vegetation of this area in recent times. Only one species of the group survived in the desert, but at the price of a dramatic adaptation: Centaurea omphalodes (Coss. & Dur.) O. Hoffm. from the oasis of Algeria is the only annual species of the Acrocentron group.
As a final argument for the predominance of the southern path of migration, in northern and middle Italy some species of C. sect. Acrocentron can be found, but they are related to the Balkan clade and did not cross the huge barrier of the Alps into France (Fig. 3). The only species of C. sect. Acrocentron present in France belongs to the widespread C. scabiosa group, with the exception of only one species shared with east Iberia (C. collina).
Migration via the southern path would also explain the pattern of distribution of C. sect. Acrocentron in Iberia. Figure 4 shows the richness in number of species in the Iberian Peninsula and Morocco. The number of taxa diminishes northwards, and in northern Iberia species of C. sect. Acrocentron are scarce and most of them (three out of four) are relictic paleopolyploids (Garcia-Jacas and Susanna, 1992).
In contrast, DIVA reconstruction suggests the Iberian Peninsula as the ancestral area for the clade comprising Iberian, North African and Sicilian species (Fig. 1). This solution seems unlikely because it contradicts biogeographic evidence: if taxa of Acrocentron migrated from Anatolia to the Aegean and the south Balkans, the easiest way to the Iberian Peninsula is the southern route via Sicily and North Africa, where landmark species exist (Fig. 3). When interpreting the DIVA results, we have to be aware that the DIVA method does not take in account geographic connections between the areas defined for the analysis (Ronquist, 1996, 1997).
A later entry of C. sect. Acrocentron following a northern pathway can be tracked in the Iberian Peninsula, originating among the species of the Balkan clade. The continuous distribution of species of this group in Europe and Middle Asia, from the Iberian Peninsula and Great Britain to Kazakhstan, should be considered a recent radiation based on a single species, C. scabiosa. Many local variants of this species have been described as new taxa on insufficient morphological basis, which confirms that they are of very recent age. Even those taxa from the C. scabiosa complex that are usually accepted by synantherologists, C. cephalariifolia from Spain (a tetraploid), C. glehnii Trautv. from Armenia and C. kotschyana Heuff. from Poland, have been diversely considered at subspecific rank within C. scabiosa. This extreme geographic radiation (Fig. 3) was favoured by the occurrence of a cold steppe continuum from Iberia to Asia during the repeated interglacial periods in the Pleistocene–Holocene (Prentice et al., 2000).
In our molecular reconstruction based on the ITS plus ETS regions, the phylogeny of the Iberian and African taxa is obscure and unresolved because the present distribution of ribotypes in Iberia is significantly mediated by the troublesome climatic history of the Iberian Peninsula and North Africa during the Holocene. The succession of glaciations forced mountain species of Acrocentron to migrate to lower latitudes (Peñalba et al., 1997), and species that were isolated became sympatric. Iberian species could migrate south to Africa during glacial maxima, and African taxa moved to the north during the warmer interglacial periods (Fig. 3). The resulting pattern of generalized introgression is reflected in the nrDNA phylogeny, with even north-Iberian cloned sequences deeply nested in the North African clade. Perhaps the most striking result, which confirms the impact of glaciation-mediated migrations and hybridization in C. sect. Acrocentron, is the otherwise implausible group formed by one cloned sequence of C. argecillensis from central Spain and another of C. litardierei from the High Atlas mountain range (Figs 1 and and22).
Our study demonstrates the usefulness of the analysis of the ETS region for phylogenetic and biogeographic investigation of highly hybridized groups. The unravelling of the pathways followed by the Acrocentron group offers a possible model for many other groups with a suggested eastern Mediterranean origin and secondary centres of speciation in the Iberian Peninsula and North Africa.
Financial support from the Spanish Ministery of Education and Science (Projects CGL2004–04563-C02-01/BOS and CGL2006–01765/BOS) and the Generalitat de Catalunya (‘Ajuts a Grups de Recerca Consolidats’ 2005/SGR/00344) is gratefuly acknowledged. The authors thanks the curators of the institutions and the collectors listed in Table 1 for kindly supplying material of some species. Suggestions by two anonymous referees for improving the manuscript are gratefully acknowledged.