Phylogeography above the species level for perennial species in a composite genus

Phylogeography above the species level is a powerful tool for investigating patterns and processes at the boundary between divergent and reticulate relationships. We examined the evolutionary history of perennial species in the western Mediterranean composite genus Helminthotheca using DNA sequence and fingerprint data. The origin of the group was in western North Africa, a region of intensive Pleistocene speciation. From here it expanded to the Iberian Peninsula and Sicily. The inferred evolutionary history is compatible with the concept of ecogeographic isolation, which refers to the fact that geographic ranges of diverging lineages are largely non-overlapping due to adaptive differentiation.


Introduction
Intraspecific phylogeography as originally defined by Avise et al. (1987) is concerned with the geographic distribution of genealogical lineages at the level of conspecific populations. Originally based upon animal mitochondrial DNA sequence data, phylogeographic studies today make use of a variety of molecular techniques and are based upon DNA sequence and fingerprint data from the mitochondrial, plastid and nuclear genomes. The field of phylogeography today extends from the level of conspecific populations to the supraspecific level (Avise 2000) and targets organisms from the bacteria and eukaryote domains of life as well as viruses. Phylogeography of closely related species spans the boundary between reticulate and divergent relationships, where a partially braided collection of allelic pathways of interbreeding individuals bifurcates into two such collections (Avise 2000). Likewise, genealogical data at the population level of closely related species are especially helpful for explaining distributions resulting from historical and ecological factors.
A region that offered many opportunities for ecogeographic and effective geographic isolation (Sobel et al. 2010) due to its variegated geologic and climatic history since the Miocene is the western Mediterranean region. The formation of the Betic-Rif orogen led to the desiccation of the Mediterranean Sea between 6 and 5.3 million years ago (mya; Hsü et al. 1973), enabling the migration of animals and plants between the two continents (Bocquet et al. 1978). At the end of the Messinian, the Strait of Gibraltar opened and water from the Atlantic Ocean refilled the dry Mediterranean Basin (Lonergan and White 1997), thus again disconnecting north-western Africa from south-western Europe.
Major climatic events after the opening of the Strait of Gibraltar were the advent of the Mediterranean climate and Pleistocene climatic oscillations. In northern Africa, the appearance of a local or temporary dry season was already recognizable in the floristic composition of the Miocene and this climate trend was accentuated in the Pliocene (Quézel 1978). In the north-western Mediterranean area, a Mediterranean climatic rhythm with summer drought appeared in the Pliocene, at 3.2 mya (Suc 1984). In the Pleistocene, the water level of the Mediterranean Sea underwent great oscillations. During the last glacial maximum (0.02 mya), it was 120 m lower than today, so that the oceanic barrier between northwestern Africa and south-western Europe was much reduced in extension, though not completely disrupted (Thiede 1978). The western Mediterranean region was an important refugium for temperate as well as Mediterranean plants and animals during the Pleistocene glacial periods (Mé dail and Diadema 2009). It is supposed that the changing environmental conditions and appearance/disappearance of biogeographic barriers have massively triggered plant and animal diversification in the Pliocene and Pleistocene (Fiz-Palacios and Valcá rcel 2013).
The genus Helminthotheca (Asteraceae, Cichorieae, Hypochaeridinae) has a primarily western Mediterranean distribution. It has been separated from Picris at the generic level on the basis of the conspicuously enlarged outer involucral bracts and comprises the annual, heterocarpic species H. echioides and H. balansae and the perennial, homocarpic species H. aculeata, H. comosa and H. glomerata (Picris comosa auct. alg.; Lack 1974Lack , 1975. Helminthotheca aculeata has another subspecies, H. aculeata subsp. maroccana, in addition to the nominal subspecies. Similarly, H. comosa also has another subspecies, H. comosa subsp. lusitanica, in addition to the nominal subspecies. The most recent common ancestor (MRCA) of extant taxa of Helminthotheca can be traced back to the Pliocene or Pleistocene (Tremetsberger et al. 2013). It is thus likely that populations have been affected by climatic changes occurring in the Mediterranean region in this epoch, possibly leading to range shifts, disruptions of previously continuous ranges and secondary contacts. This study aims at shedding light on the evolutionary history of the perennial species of Helminthotheca. Using molecular makers from the nuclear and plastid genomes, we addressed the following questions: (i) Does genetic grouping correlate with the present taxonomic treatment? (ii) Where did perennial species of Helminthotheca originate? (iii) Along which routes did they spread to occupy their present areas? The inferred evolutionary history is discussed with respect to the evolutionary forces that might have been relevant for speciation.

Plant material
We sampled populations belonging to subspecies of Helminthotheca aculeata and H. comosa in Algeria, Italy (Sicily), Morocco, Portugal and Spain (Table 1, Fig. 1). Helminthotheca echioides was included as outgroup. Vouchers of all populations used are deposited in the herbaria at Seville (SEV) and/or Vienna (WU). The Algerian species H. balansae and H. glomerata (Qué zel and Santa 1963) could not be sampled for this study. In the field, leaves from individual plants were conserved in silica gel. Distribution ranges of the taxa under study were compiled from survey of material in the herbaria at Barcelona (BC), Geneva (G), Madrid (MA), Montpellier (MPU), Paris (P), Salamanca (SALA) and Seville (SEV). Locations mentioned in Pignatti (1982) and Sauvage (1961) were also used. DIVA-GIS ver. 7.1.7.2 (Hijmans et al. 2001) was used to display distribution ranges and sample locations.

Amplified fragment length polymorphism (AFLP)
The protocol for extraction of genomic DNA from silica gel-dried leaves, restriction-ligation (EcoRI and MseI), preselective and selective amplification, and electrophoresis on a capillary sequencer (3130xl Genetic Analyzer, Applied Biosystems) exactly followed Tremetsberger et al. (2009). Three selective primer combinations [EcoRI(Fam)-ACA/MseI-CAT, EcoRI(Vic)-AGG/MseI-CTCand EcoRI(Ned)-AAC/MseI-CTT] were selected following a primer trial testing 12 selective primer combinations, based on the criteria of interpretability (presence/absence), spread of bands in the region 50 -500 bp, and number of bands ( 70 -90 per primer combination). Raw data were aligned with the GeneScan 500 ROX size standard using ABI PRISM GeneScan ver. 3.71 (&Applied Biosystems). The presence/absence of bands in all individuals and replicates was scored with Genographer ver. 1.6.0 (Benham 2001) after normalizing on total signal in one file. We scored 96 individuals from 25 populations of the perennial species of Helminthotheca. Twenty-two individuals were replicated from restriction-ligation and used to calculate the error rate as the ratio between observed number of phenotypic differences and total number of phenotypic comparisons between replicates (Bonin et al. 2004). Three individuals of H. echioides were used as outgroup. Only loci observed in the ingroup were scored in the outgroup. We employed split decomposition (Bandelt and Dress 1992) for constructing a phylogenetic network, which is preferred over a strictly bifurcating tree at the population level, where gene flow is to be expected. FAMD ver. 1.108 (Schlü ter and Harris 2006) was used to calculate a matrix of Nei and Li (1979) distances with the length of the restriction enzyme's recognition sequence set to six between all pairs of individuals. This matrix was imported into SplitsTree ver. 4.8 (Huson and Bryant 2006) to construct a NeighborNet network. Splits, whose weight did not exceed a threshold of 0.001, were removed.
Bayesian admixture clustering (Corander et al. 2003;Corander and Marttinen 2006) was used to estimate the number of genetically coherent clusters in the perennial species of Helminthotheca with BAPS ver. 5.2 (Corander et al. 2004(Corander et al. , 2008. The maximum number of populations for mixture clustering was set to 20 (5×), 25 (5×) and 30 (5×). For admixture clustering based on results of mixture clustering, the minimum size of a population was set to 1, the number of iterations to 200 and the number of reference individuals to 200 with 20 iterations each.
In order to determine which taxon might be ancestral, we generated a phylogenetic tree, which also included three individuals of the outgroup, H. echioides, by neighbour-joining analysis based on Nei-Li distances in PAUP* ver. 4.0b10 (Swofford 2003). The support of nodes was estimated by 500 bootstrap replicates. Clusters containing only individuals of the same population or taxon were collapsed using FigTree ver. 1.3.1 (Rambaut 2006(Rambaut -2009. In order to distinguish between incongruent species trees, we adopted the approach of Bryant et al. (2012) to infer species trees from AFLP data in the framework of the multispecies coalescent, implemented in SNAPP ('SNP and AFLP Phylogenies'), a Bayesian MCMC sampler, which interfaces with the BEAST package (Drummond and Rambaut 2007 (Bouckaert et al. 2002(Bouckaert et al. -2014. Trees and associated parameter values were logged every 1000 states. Results of the MCMC runs were analysed with Tracer ver. 1.5.0 (Rambaut andDrummond 2003 -2009), discarding the first 10 % as burn-in. Trees created in the two MCMC runs were combined with LogCombiner ver. 2.1.3, again discarding the first 10 % of each run as burn-in. A species tree was drawn from the combined trees with DensiTree ver. 2.1.11 (Bouckaert 2010). Within-population genetic diversity was calculated as 'average gene diversity over loci' in Arlequin ver. 3.5.1.2 (Excoffier et al. 2005; only for populations with a minimum of three individuals). As indicators of divergence due to long-lasting isolation of populations, we estimated the number of private AFLP bands in a population (i.e. those bands occurring exclusively in the respective population) and the rarity index. Only populations with a minimum of two individuals were used. Furthermore, some populations with more than two individuals were also omitted, so that a similar number of populations was included for each taxon. The number of private bands was counted with FAMD ver. 1.25. The rarity index equivalent to frequency-down-weighted marker values (DW) of Schö nswetter and Tribsch (2005) was calculated as 'rarity 1' using the R-script AFLPdat (Ehrich 2006) in R ver. 2.11.1 (&The R Foundation for Statistical Computing). For each individual, each AFLP band was divided by the total number of occurrences of this band in the dataset. These relative values were then summed to the rarity index for this particular individual. Population values were calculated as averages of individual values.
Forward and reverse sequences were assembled with SeqMan II ver. 5.05 (DNASTAR, Inc., Madison, WI, USA). The sequences were aligned manually in BioEdit ver. 7.0.5.3 (Hall 1999). SeqScape ver. 2.5.0 (Applied Biosystems) was used to check all mutations in the alignments as well as multiple states (additivities) found in many ITS sequences. Because the peaks of the two bases contained in the multiple states had similar heights, they were interpreted as real polymorphisms due to incomplete concerted evolution and not as sequencing uncertainties. BioEdit was also used to calculate G + C contents of ITS sequences. We also reconstructed the secondary structure of ITS2 of the perennial species of Helminthotheca, using the RNA folding form of mfold ver. 3.5 (Zuker 2003), based on the common structure described for Asteraceae by Goertzen et al. (2003). Sequences were deposited in the EMBL Nucleotide Sequence Database (Stoesser et al. 2003; Table 1).
For phylogenetic analysis of ITS sequences, we adopted the maximum likelihood (ML) approach using the RAxML software on the Vital IT unit of the Swiss Institute of Bioinformatics Web server (http://embnet.vital-it.ch/raxmlbb/index.php, 24 July 2015; Stamatakis et al. 2008), with individual general time reversible (GTR) models of nucleotide substitution with gamma-distributed site-to-site rate variation and a proportion of invariable sites being estimated for the three partitions ITS1, 5.8S rDNA and ITS2. Support values were obtained by bootstrapping (100 replicates).
For analysis of concatenated ndhF-rpl32 and rpl32-trnL sequences, the maximum parsimony (MP) approach was adopted due to its ease of incorporating information contained in indels, which were coded using the simple indel coding method implemented in SeqState ver. 1.3.2 (Mü ller 2005). A heuristic search under the MP criterion was carried out with 1000 random-addition-sequence replicates and TBR swapping with no more than 10 trees of score ≥10 retained in each replicate with PAUP* ver. 4.0b10 (Swofford 2003). Bootstrap support was estimated with 500 bootstrap replicates (Felsenstein 1985; heuristic search with 10 random-addition-sequence replicates and TBR swapping with no more than 10 trees of score ≥10 retained in each replicate).
A species tree was constructed from ITS and plastid DNA sequences (without indel coding) in BEAST ver. 2.1.3 using the *BEAST template (Heled and Drummond 2010), with a Jukes-Cantor (JC69) site model, a relaxed log normal clock model, a birth-death tree model with constant population size and 10 million states of the MCMC chain. Individuals of putative hybrid origin (see below) were not included. Populations were assigned to the following taxa: H. aculeata subsp. aculeata, A1-A5; H. aculeata subsp. maroccana, M1 -M3 (the plastid DNA sequences of two individuals that have the 'comosa/ lusitanica'-plastid type [populations M1 and M3; see below] have not been included); H. comosa subsp. comosa, C1-C6; H. comosa subsp. lusitanica, S1-S8; and H. echioides, E1-E5. Use of other models (e.g. the GTR site model) did not affect the result. Trees were displayed in DensiTree ver. 2.1.11 (10 % burn-in).

Molecular clock
We adopted the approach of Tremetsberger et al. (2013) for dating the age of the perennial species of Helminthotheca. The ITS1 and ITS2 alignments of Cichorieae (Tremetsberger et al. 2013; available from TreeBASE) and of Helminthotheca were merged. To simplify the calculation, some sequences were removed; totals of retained sequences were 25 for Helminthotheca and 46 for Cichorieae and outgroup taxa. BEAST ver. 1.6.2 was used as outlined in Tremetsberger et al. (2013) with stem group node calibration. Results of four independent MCMC runs were analysed with Tracer ver. 1.5.0 and combined with Log-Combiner ver. 1.6.2 (10 % burn-in). The maximum clade credibility tree was obtained using TreeAnnotator ver. 1.6.2 and graphed with FigTree ver. 1.3.1.

AFLP
In total, 231 AFLP bands in the range of 90 -430 bp were scored for the 96 individuals of the perennial species of Helminthotheca. Of these, 207 (i.e. 90 %) are polymorphic. All individuals have unique AFLP phenotypes. The overall error rate was 4 % (ranging from 0 to 41 % for individual loci). Removal of seven loci with individual error rates .25 % did not significantly alter the phylogenetic network ( Fig. 2A). We thus decided to rely on the original matrix for all further calculations.
In the Bayesian mixture clustering analysis, the first nine best visited partitions revealed four clusters, whereas the 10th best visited partition revealed five clusters. The values for the log of the marginal likelihood of the nine optimal partitions range from 25916 to 25960 (the 10th best visited partition has a value of 25971). The posterior probability for K ¼ 4 is 1. The four clusters revealed (Figs 1 and 2) are (i) populations A1 -A5 of H. aculeata subsp. aculeata (Algeria, Sicily), (ii) populations M1 -M3 of H. aculeata subsp. maroccana together with populations AM1 and AM2, which morphologically resemble H. aculeata subsp. aculeata, and population CM, which morphologically resembles H. comosa subsp. comosa (all Morocco), (iii) populations C2 -C6 of H. comosa subsp. comosa (Spain), and (iv) populations S1-S8 of H. comosa subsp. lusitanica (Portugal, Spain) together with population CS (Spain). Some evidence for admixture was also found, namely in some individuals of populations CM and CS, which show admixture from the H. comosa-cluster (Fig. 2B).
Evaluation of sampled trees from the coalescent approach to species tree reconstruction revealed an effective sample size (ESS) of .10 000 for the two combined runs. The traces did not show signs of nonconvergence. All clades have posterior probabilities of 100 %, yielding a single consensus tree (Fig. 3A). In a tree lacking short branches, results of analyses with and without correction for dominance do not differ (Bryant et al. 2012), so that we relied only on the analysis without correction for dominance. The sequence of divergence suggested by the obtained species tree is: first, divergence of an Afro-Sicilian and an Afro-Iberian lineage; second, divergence of H. aculeata subsp. maroccana and H. aculeata subsp. aculeata in the Afro-Sicilian lineage; and third, divergence of H. comosa subsp. comosa and H. comosa subsp. lusitanica in the Afro-Iberian lineage.
Accumulation of private and rare AFLP bands in populations due to mutations indicates long-term in situ persistence of populations. This is in contrast to newly established populations, which should have fewer such bands. The mean number of private bands and the rarity index are slightly higher in populations of the Afro-Sicilian lineage than in populations of the Afro-Iberian lineage ( Table 2). This trend is accentuated, when the estimates are calculated for taxa rather than for populations. Helminthotheca aculeata subsp. aculeata

DNA sequences
The G+C content of ITS1 and ITS2 sequences is very similar among accessions of the perennial species of Helminthotheca (mean ¼ 53.3 %; SD ¼ 0.2 %) and H. echioides (mean ¼ 54.0 %; SD ¼ 0.3 %), suggesting the absence of pseudogenes. The ITS2 secondary structure is common to all individuals of Helminthotheca analysed [see Supporting Information- Fig. S1]. It is also well in accordance with the ITS2 secondary structure of Asteraceae presented by Goertzen et al. (2003). No compensatory base changes (Coleman 2007) exist among individuals of Helminthotheca, neither when comparisons are made among perennial species nor when these are compared with H. echioides. The ITS tree (Fig. 4A) reveals two groups. Sequences of populations M1-M3 are in basal position together with one sequence of population AM1 and one sequence of population CM. All remaining sequences are in a derived group with 72 % bootstrap support. The plastid DNA tree (Fig. 4B) also reveals two main groups. One group with 79 % bootstrap support contains sequences of the Afro-Sicilian AFLP group (all but two sequences of populations M1-M3 and all sequences of populations AM1, AM2, and A1-A5). The other group with 81 % bootstrap support contains sequences of the Iberian AFLP group (populations C1-C6, CS and S1-S8) together with one sequence of population M1, one sequence of population M3 and sequences of population CM.

Hybrids
Differences in the individual trees derived from the different datasets allow identifying individuals of hybrid origin. Individuals of population CM largely correspond to the 'maroccana'-AFLP group (with two individuals showing admixture from the 'comosa'-AFLP group), one individual has the more basal 'maroccana'and the other the more derived 'aculeata/comosa/ lusitanica'-ribotype, but all individuals have the 'comosa/lusitanica'-plastid type. Similarly, one individual of population M1 and one individual of population M3 also have the 'comosa/lusitanica'-plastid type. One individual of population AM1 and both individuals of population AM2 morphologically resemble H. aculeata and have the 'aculeata/comosa/lusitanica'-ribotype, but correspond to the 'maroccana'-AFLP group. Finally, some individuals of population CS show evidence of admixture by Bayesian admixture clustering of AFLP AoB PLANTS www.aobplants.oxfordjournals.org data. When individuals of putative hybrid origin are excluded, Bayesian inference results in a best-scoring species tree derived from ITS and plastid DNA sequences (Fig. 3B) that is congruent with the species tree derived from AFLP data.

Ancestral area and age estimation
Model-based biogeographic reconstruction of ancestral areas (Fig. 5) gives an idea of the biogeographic history of the perennial species of Helminthotheca, albeit with little confidence. Western North Africa with its mountain ranges Rif and Middle Atlas is revealed as the ancestral area of the entire clade as well as of the common ancestor of H. aculeata subsp. aculeata and H. aculeata subsp. maroccana. The common ancestor of H. comosa subsp. comosa and H. comosa subsp. lusitanica is inferred to have lived in an area spanning W North Africa, the W Betic region and the W Iberian region. Molecular clock analysis reveals that the minimum age of the MRCA of the perennial clade dates well to the middle of the Pleistocene (median ¼ 1.09 mya; 95 % HPD interval ¼ 0.48 -1.93 mya) [see Supporting Information- Fig. S2].

Specific entities
Genomic variation as evidenced by Bayesian mixture clustering of AFLP data (Fig. 2B) shows a basic concordance with morphologicalvariation(S.Talaveraetal.,unpubl.data).Diagnostic characters for distinguishing H. comosa subsp. lusitanica from the nominal subspecies are provided by the involucral bracts of the heads. In the former, the subapical appendix of the inner involucral bracts is shorter (1-3.5 mm long in H. comosa subsp. lusitanica vs. 5-8 mm long in H. comosa subsp. comosa) and projects less above the bracts   (Ree and Smith 2008). Global 2ln L at root node is 9.405, the estimated dispersal rate is 1.124, and the estimated local extinction rate is 4.285e209. (A) Area adjacency matrix used for input. (B) Species tree (consensus tree of *BEAST analysis) displaying most probable ancestral areas. (C) Splits with associated log likelihoods (ln L) and relative probabilities (Rel. Prob.) for each node. Following De Queiroz (2007) in treating the existence as separately evolving segments of lineages as the only necessary property of species, we therefore propose to distinguish H. comosa subsp. lusitanica as a separate species from H. comosa and H. aculeata subsp. maroccana as a separate species from H. aculeata. As outlined above, lines of evidence for lineage separation come from AFLP data and morphology. The lineages also have distinct geographic distributions (Fig. 1) and ecological preferences (see below). In this sense, we will refer to H. comosa subsp. lusitanica as H. spinosa, comb. nov., and to H. aculeata subsp. maroccana as H. maroccana, comb. nov., in the following (the new combinations are presented in the section Taxonomic Implications).

North-western African origin
Although ancestral area reconstruction is not quite conclusive, it suggests that the perennial clade of Helminthotheca originated in western North Africa, possibly in the Middle Atlas or Rif. A caveat of the biogeographic analysis, however, is that few populations from central North Africa (Tell Atlas; especially H. glomerata) and no populations from the High Atlas and Anti-Atlas have been included, so that we cannot more precisely determine the ancestral area.
Our results for the perennial clade of Helminthotheca are well in line with observations in the closely related Hypochaeris sect. Hypochaeris ). Both groups originated in western North Africa in the Pleistocene or Pliocene; their origins were thus associated with the ascent of the Mediterranean-type climate and/or Pleistocene climatic oscillations. The importance of the High and Middle Atlas region in addition to the Betic-Rif region for the evolution of Mediterranean plant taxa has been suggested by Mé dail and Qué zel (1997). The High and Middle Atlas Mountains, just as the Rif, are known to have a high rate of endemism, which is explained by the long isolation and high altitude of these massifs, which made them serve as refugium for the Tertiary residual flora. In addition, there is also a large number of schizoendemics that demonstrate the role of the Atlas Mountains in more recent speciation processes (Mé dail and Qué zel 1997). This is in line with climatic reconstructions, which suggest that a Mediterraneantype climate with summer drought was present in northern Africa earlier than in southern Europe (Qué zel 1978;Suc 1984). Fauquette et al. (1999) describe Pliocene vegetation assemblages resembling the modern thermo-Mediterranean formation for the regions Andalusia (southern Spain), North Africa (Morocco, Algeria, Tunisia) and Sicily, corresponding to those regions inhabited by perennial species of Helminthotheca. In accordance with our age estimates, we might hypothesize that after the common ancestor of the perennial clade of Helminthotheca had originated in response to the ascent of the Mediterranean-type climate in northern Africa, it held on during the Pliocene and diversified in the Pleistocene.

Migratory routes out of western North Africa
The evolutionary scenario derived from our data suggests two migratory routes out of western North Africa and across the Mediterranean Sea. The first expansion was to the north, across the Strait of Gibraltar, to the Iberian Peninsula. The second expansion was to the east, along central North African territories, to Sicily and the Apennine Peninsula.
Numerous studies have investigated genetic differentiation between populations on both sides of the Strait of Gibraltar (e.g. Arroyo et al. 2008;Escudero et al. 2008), with differing outcomes. Whereas the Strait of Gibraltar has been an effective barrier to gene flow in Carex helodes after its dispersal from the Iberian Peninsula to North Africa (Escudero et al. 2008), extensive gene flow across the Strait of Gibraltar via seed dispersal has been documented in Calicotome villosa (Arroyo et al. 2008). In the central Mediterranean region, the Strait of Sicily separating North Africa and Sicily has also functioned as an important route of dispersal for plants and animals (e.g. Habel et al. 2008;Ferná ndez-Mazuecos and Vargas 2011). Dispersal was probably facilitated by lowered sea water levels during the Pleistocene glacial periods (Thiede 1978). In Helminthotheca, seed dispersal could have been effectuated by wind or migratory birds (Finlayson 1992).

Ecogeographic isolation
As in other Mediterranean plant groups of Pliocene/Pleistocene age such as Antirrhinum (Vargas et al. 2009) and Reseda sect. Glaucoreseda (Martín-Bravo et al. 2010), there is also a clear geographic component to speciation in the perennial clade of Helminthotheca. The biogeographic analysis suggests two lineage divergence events within western North Africa, the first yielding an Afro-Sicilian and an Afro-Iberian lineage and the second yielding H. maroccana (as progenitor) and H. aculeata (as derivative) within the Afro-Sicilian lineage. The placement of H. glomerata in this scenario still has to be determined. The third lineage divergence event is situated in the western Iberian region, yielding H. comosa (as progenitor) and H. spinosa (as derivative).
Accurate identification of species pairs is crucial for investigating isolating barriers. The limit between the distributions of H. comosa and H. spinosa closely follows the Guadalquivir river basin, which separates the Mesozoic and Cenozoic Betic Cordillera ('Western Betic Region'; García-Barros et al. 2002) inhabited by H. comosa and AoB PLANTS www.aobplants.oxfordjournals.org the Paleozoic Iberian Massif ('Western Iberia') inhabited by H. spinosa. Characteristics of the two regions such as those associated with the substrate (calcareous in the Betic Cordillera, acidic in the Iberian Massif) are expected to exert divergent selective pressure on populations, resulting in ecogeographic isolation of the diverging lineages (Sobel et al. 2010). Ecogeographic isolation might also be hypothesized for the species pair H. aculeata/H. maroccana, in response to differences in substrate or precipitation. Helminthotheca aculeata grows on calcareous substrate of the Cenozoic, whereas H. maroccana grows on crystalline rocks and schist. Hybrids exist in contact zones between the regions. For example, some populations of H. comosa grow in calcareous enclaves, which exist in the western Iberian region. Hybrids between H. spinosa and H. comosa are found precisely in the ecotones between acidic and calcareous soils.
As already supposed by Sauvage (1961), assessment of ploidy levels could shed more light on the issue of hybridization coupled with polyploidization in Helminthotheca.

Taxonomic implications
As a result of the genetic analyses and the morphological survey of the populations included in this work (S. Talavera et al., unpubl. data), two new combinations and descriptions of three hybrids that are new to science are now necessary. plant material, Ali Loukkas and Ali Mahmoudi, former directors of National Parks of Theniet-el-Had and Gouraya (Algeria), for their help in the field and courtesy for collecting fresh material, Eric Chambers (Vienna) for proofreading and two anonymous referees for constructive criticism.

Supporting Information
The following additional information is available in the online version of this article - Figure S1. Common secondary structure of ITS2 rRNA of Helminthotheca. Figure S2. Chronogram of Helminthotheca and the two closest outgroup genera, Leontodon and Picris, based on ITS1 and ITS2 sequences.

Literature Cited
Arroyo J, Aparicio A, Albaladejo RG, Muñ oz J, Braza R. 2008. Genetic structure and population differentiation of the Mediterranean pioneer spiny broom Calicotome villosa across the Strait of Gibraltar. Biological Journal of the Linnean Society 93:39 -51.