The analysis of three combined nuclear DNA sequence sets for all diploid Hordeum species resulted for the first time in a phylogenetic tree with strongly supported backbone structures and resolution within the New World taxa of the genus. In general, statistical support of the clades (posterior probabilities and bootstrap values) was higher than in earlier separate analyses of the same loci (Komatsuda et al., 1999; Blattner, 2004) or in the combination of EF-G, DMC1, and chloroplast data (Petersen & Seberg, 2003). The general topology of the tree is entirely compatible with the postulated genome relationships (Fig. 3a) and earlier cytological (for references see von Bothmer et al., 1995), amplified fragment length polymorphism (AFLP) (El-Rabey et al., 2002), EF-G (Komatsuda et al., 1999, 2002), and ITS analyses (Blattner, 2004). Differences between the topology observed in the present study and the results of a separate analysis of the DMC1 locus alone (data not shown) concern only H. brevisubulatum and H. marinum, which group together with the basal New World taxon H. brachyantherum ssp. californicum, although with low statistical support. As this DMC1 topology contradicts the relationships deduced from independent nuclear data sets (see earlier), these differences are probably caused by the persistence of ancient polymorphisms and incomplete lineage sorting among DMC1 alleles in the young H-genome group (Maddison, 1997; Ting et al., 2000; Petersen & Seberg, 2003). Chloroplast data produced pronouncedly different tree topologies (Doebley et al., 1992; Nishikawa et al., 2002; Petersen & Seberg, 2003) when compared with the nuclear data. These incompatibilities were the reason for omitting chloroplast sequences from the combined data analysis here. The inconsistencies can mostly be attributed to lateral chloroplast transfer among some Hordeum taxa (Nishikawa et al., 2002) and particularly incomplete ancient lineage sorting (S. S. Jakob & F. R. Blattner, unpublished).
The occurrence of two ITS paralogs in the New World Hordeum clade and the reciprocal loss of one or the other paralog in some species results in seemingly contradicting phylogenetic topologies in this species group (Blattner, 2004) when analysed in a single data set (Fig. 1). Combining both ITS paralogs as separate loci with DMC1 and EF-G in a data matrix and coding the missing sequences as unknown character states resulted in a consensus tree with low resolution within the South American taxon group (data not shown). These problems were overcome by a separate analysis of data matrices containing only single ITS types and an integration of the phylogenetic information with an MRP supertree analysis. Supertree approaches are often used to combine data sets with differing taxon samples (Bininda-Emonds et al., 2002). In the case of bidirectional homogenization resulting in the loss of paralogous gene copies from several species, no possibility exists to recover the lost alleles. Phylogenetic analysis thus needs to combine the information from the separate loci (gene or locus trees) to arrive at the species’ phylogenies (Baum, 1992). The use of the recoded strict consensus trees from both data sets in MRP resulted in c. 250 000 most parsimonious trees, separating H. brachyantherum from the South American taxa, but with low resolution in the latter clade (data not shown). As this might be the result of under-representation of the data matrix (i.e. small number of matrix elements), all single most parsimonious trees instead of the consensus trees were recoded in the MRP matrix (Baum, 1992; Bininda-Emonds & Bryant, 1998). The analysis of these data substantially improved the resolution of the internal structure of the South American species (Fig. 2). The involvement of identical sequences for the Old World taxa in both data subsets resulted in pseudoreplication of these data in the MRP analysis (Bininda-Emonds, 2004). However, as only the relative phylogenetic positions of derived taxa with a single ITS paralog were of interest here, pseudoreplication of primary data in all other basal taxa is irrelevant for the resulting tree.
Hordeum consists of two major clades: (1) the primarily western Asian and Mediterranean species of the I- and Xu-genome groups, and (2) a clade formed by Eurasian H. marinum (Xa genome) and the large group of H-genome taxa (Fig. 3a). The basal branches in Hordeum are statistically well-supported, with the exception of the position of H. murinum, which in some analyses received only low bootstrap support (Fig. S1). However, in contrast to the separate analysis of ITS sequences (Blattner, 2004) neither of the analyses done here resulted in a sister-group relationship of H. murinum with the Xa/H-genome clade. In the H-genome clade the clear separation of the exclusively Asian taxa from the remaining New World taxa was highly supported (= 97% bs, 1.0 pp).
Within the closely related New World clade of the H-genome taxa the combination of nuclear DMC1, EF-G, and ITS sequences (Fig. 1), together with the use of an MRP supertree approach (Fig. 2) for the first time resulted in a relatively clear phylogenetic pattern. The North American H. brachyantherum was sister to the primarily South American species. Within this latter group H. chilense is sister to two geographically circumscribed clades, consisting of three southern species (H. comosum, H. patagonicum, and H. pubiflorum) and a northern group, respectively. The four mostly northern Argentinean species (Fig. 3a) group together with two taxa from North America. This clade, however, received only low statistical support (0.7 pp), although the two South American–North American species pairs within this group were well-supported (= 0.94 pp). The MRP supertree analysis also grouped H. cordobense and H. muticum with this northern clade, further supporting the geographical partition of the South American Hordeum species, as both these species occur in northern Argentina and H. muticum even reaches the Andes of Ecuador. This northern species group overlaps with section Anisolepis of von Bothmer et al. (1995), albeit with two differences: It includes H. erectifolium (section Stenostachys sensu von Bothmer et al., 1995) but excludes the basal South American H. chilense, as this species is sister to both the northern and the southern species group.
Internal transcribed spacer (Blattner, 2004), chloroplast (Nishikawa et al., 2002) and cytological analyses (Linde-Laursen et al., 1992) defined Asian H. roshevitzii and North American H. brachyantherum (or a close extinct relative) as progenitors of tetraploid H. jubatum. These data also support the close relationship of H. guatemalense and H. jubatum, with ITS data further indicating H. intercedens as a parental taxon involved in the formation of H. guatemalense (Blattner, 2004). The close relationship of H. capense and H. secalinum was long recognized. Some authors even treated these species as conspecific (von Bothmer et al., 1995; Baum & Johnson, 2003). The contribution of H. marinum as one parental taxon has been shown in several analyses (Doebley et al., 1992; Nishikawa et al., 2002; Baum & Johnson, 2003; Blattner, 2004; Petersen & Seberg, 2004), but contradictory hypotheses about the second parent were proposed based on chloroplast data (H. brachyantherum; Doebley et al., 1992), DMC1 (H. brevisubulatum; Petersen & Seberg, 2004), and ITS sequences (a basal taxon from the South American H-genome group; Blattner, 2004). A South American origin (close to H. muticum) of one parent of H. capense was postulated from the analysis of 5S rDNA loci (Baum & Johnson, 2003). However, these authors interpreted the absence of a specific 5S rDNA unit from H. secalinum as indication of only a distant relationship to H. capense, despite high sequence similarity at shared 5S rDNA loci. Owing to the dynamic nature of 5S rDNA loci (Schubert & Wobus, 1985; De Bustos et al., 1996; Taketa et al., 1999, 2001), the absence of a locus does not necessarily contradict a close relationship of the two species, as number of loci is variable even within narrow species groups. The combined data matrix for all species here supported the close relationship between H. capense and H. secalinum and suggested H. marinum and an H-genome species (Fig. S3) as progenitors. However, no decision about an Old or New World origin is possible based on the data presented here. The combination of ITS and EF-G data clearly supports a South American species as one progenitor of these taxa, but DMC1 data are inconsistent with a New World origin of the H-genome parent of both species (data not shown). As the DMC1 sequences of the tetraploids fall into the already contradictory marinum/brevisubulatum-clade (see above), no safe conclusion about the phylogenetic position of the H-genome parent is possible from this locus and the combined data. In the phylogenetic tree in Fig. 3a the ambiguity over the origin of the H genome is depicted by dashed lines.
The congruence among trees based on different nuclear data sets and phylogenetic methods and the statistical support of the diploid backbone phylogeny of Hordeum provides a sound phylogenetic framework for biogeographic analysis. The different tree topologies resulting from bidirectional homogenization of two ITS paralogs (Blattner, 2004) concern exclusively South American lineages and thus do not influence the interpretation of the general biogeographical pattern of the genus. Dispersal–vicariance analysis (Ronquist, 1996, 1997) was used to estimate ancient distribution areas for the nodes in the Hordeum phylogeny (Fig. S4 and Table S1). One characteristic of DIVA is that it adds no costs for the loss of character states (i.e. the extinction of taxa in specific areas). Thus, it results in an increasing number of possible distribution areas towards the basal nodes in a tree when no constraints for the maximum number of distribution areas are imposed. Moreover, DIVA does not take into account the relative or absolute nodal ages in the tree and the relative geographical positions of the specified distribution areas. This, too, adds to the loss of meaningful information for the deeper nodes in a tree. DIVA provides, for example, an optimization of the basal nodes to the occurrence of Hordeum in all distribution areas, colonized by the genus today (Table S1). This is a very unlikely scenario with regard to continental positions 12 MYA. As DIVA provides the technically possible character optimizations for the nodes in a tree, the consideration of all additional information helps to exclude the unlikely character states. The biogeographic scenario for Hordeum developed below is thus only one of several possibilities provided by DIVA. It is, however, parsimonious as it prefers easily crossable landbridges over intercontinental long-distance dispersals, and includes information on age estimates to reconstruct the sequence of migration and intercontinental colonization.
The oldest diploid Hordeum groups occur in South-west Asia, thus pointing to this region as the probable center of origin of the genus (I; roman numbers refer to the dispersals depicted in Fig. 3), although range shifts of the taxa due to climate changes during the last 12 MY are possible. This area assignment is in accord with the proposed Eurasian origin of Triticeae (Hsiao et al., 1999). From South-west Asia migration took place into the western Mediterranean and north-western Europe (H. marinum). A route to the east (H-genome clade) resulted in the colonization of Central and East Asia (II, III) and, via Beringia, of North America (IV). Diploid H. brachyantherum today occurs only in California. The autotetraploid cytotype of this species, however, is disjunctively distributed from California northward along the Pacific rim of North America to coastal Kamtchatka and in a small area in Newfoundland (von Bothmer et al., 1995), possibly indicating a formerly wider distribution of the species. From North America Hordeum reached South America (V), where H. chilense is sister to all other taxa. The phylogenetic data indicate two independent dispersals back to the northern hemisphere, resulting in the colonization of North America by the South American progenitors of H. pusillum (VI) and H. intercedens (VII). Surprisingly, the sister of the widespread North American H. pusillum is the ecologically narrow endemic H. erectifolium (von Bothmer et al., 1995). This provides another example for a major shift in ecology in plants after colonization of a new continent (Morrell et al., 2000).
The allotetraploid H. jubatum combines a North American brachyantherum-like genome with an Asian genome most similar to H. roshevitzii (Linde-Laursen et al., 1992; Blattner, 2004). The occurrence of this Asian genome in the mainly North American H. jubatum necessitates a second traverse of Beringia (VIII). The close relationship of H. guatemalense with the widespread North American H. jubatum and the southern Californian H. intercedens (Doebley et al., 1992; Nishikawa et al., 2002; Blattner, 2004) indicates a colonization of Central America from the north (IX). The formation of H. secalinum in Europe, which involved either a Central Asian or a South American taxon (X) together with H. marinum, could not be unambiguously clarified by the molecular data. Although rDNA (Baum & Johnson, 2003; Blattner, 2004) and combined EF-G/ITS data strongly support a South American origin of one of the H. secalinum/H. capense progenitors, EF-G does not provide sufficient resolution in this part of the tree, and DMC1 contradicts this hypothesis (Petersen & Seberg, 2004). Furthermore, a transatlantic dispersal from the Southern to the Northern Hemisphere is unusual and, to my knowledge, has not been documented for other plant species. Unfortunately, chloroplast data cannot contribute to the solution of this question, as H. marinum is the maternal parent (Nishikawa et al., 2002) of the allopolyploid. The close relationship of H. secalinum and H. capense indicates the colonization of the Cape Province from Europe after the formation of H. secalinum or by the common ancestor of both these species (XI).
The geographical expansion of Hordeum is mostly restricted to the H-genome clade and started c. 6 MYA, long after Gondwana and Laurasia broke up, and clearly before the spread of humans across the world. Thus, the intercontinental distribution discussed here is neither connected with the occurrence of ancient Hordeum lineages in Gondwana, as speculated by Baum & Johnson (2003), nor with recent human influences on potential habitats or direct diaspore dispersal by humans. Plant migration between Asia and North America via Beringia is not unlikely for temperate and temporarily also for subtropical taxa (Tiffney, 1985). The present distribution of H. jubatum and H. brachyantherum in Alaska and adjacent Siberia supports the existence of this migration route and, furthermore, an occasionally continuous distribution of Hordeum species across Beringia. Dating of the split between New and Old World taxa (6–5 MYA) places this event before the start of the Quaternary climatic cycles (c. 2.5 MYA), which periodically closed Beringia for temperate plants and thus probably isolated Asian from New World Hordeum populations. The split between Old World and New World taxa is therefore interpreted as a vicariance event (Fig. 3a) caused by climatic changes that separated populations and prevented gene flow across Beringia by forcing the species into more southerly habitats. Vicariance also seems possible for the separation of Central American H. guatemalense from its North American progenitors, as during the ice ages temperate to dry climatic conditions reached far more south into Central America than today.
The next step in the range expansion of Hordeum involved the colonization of southern South America, and dates back c. 4 MY. The Central American landbridge was already in place (Graham, 2003) when the genus had to cross this area on its way to the south. Thus, two scenarios can be proposed. One involves southward migration of an H. brachyantherum-like species along the western American mountain ridges, the other long-distance dispersal between California and southern Chile. Disjunctive distributions between California and Chile are not uncommon (Raven, 1963; van Heusden & Bachmann, 1992; Morrell et al., 2000; von Hagen & Kadereit, 2001) and are thought to be mostly mediated by birds migrating along the west coast of the Americas. The basal taxon in the South American clade is H. chilense, sharing its distribution in summer-dry Chilean areas with several other taxa of North American origin. The ecology and distribution areas of extant Hordeum species together with the phylogenetic data presented here support the long-distance dispersal alternative in Hordeum. In case of stepwise southward migration the basal South American taxa would be expected to occur somewhere along the way in Central America or the Andes. However, the only South American high-altitude taxon occurring as far north as Ecuador is H. muticum. This taxon belongs to the northern clade of the South American taxa and possesses only one of the two ITS types occurring in South America (Blattner, 2004). This loss of type-2 ITS is a derived character state in South American Hordeum species, making a basal position of H. muticum highly unlikely. By contrast, H. chilense possesses both ITS types, which further supports the basal position of this species in the South American clade.
The biogeographic scenario for the South American species thus would involve initial long-distance dispersal from California to Chile and an expansion of the distribution area to the east of the Andes. From there, two migration routes could be inferred by the phylogenetic data. One group spread to the south and today occurs mainly in southern Patagonia, Tierra del Fuego and the southern Andes (H. comosum, H. patagonicum and H. pubiflorum), while a second lineage reached northern Argentina and Uruguay, and colonized the northern Andes. This intracontinental migration pattern of Hordeum, starting after long-distance dispersal from Chile and reaching secondarily northern Andean habitats in Peru is paralleled by the migration route of South American Microseris pygmaea (Asteraceae), for which a similar distribution pattern was found (Lohwasser et al., 2004). In the course of range expansion and the colonization of diverse South American habitats rapid speciation in Hordeum started c. 2 MYA. This radiation possibly is related to the Quarternary climatic oscillations (Blattner, 2004).
Two independent and relatively young reintroductions (c. 1 MYA) of South American Hordeum species to North America might also have been mediated by migrating birds. Species such as the buff-breasted sandpiper (Tryngites subruficollis) or the American golden plover (Pluvialis dominica) winter in Argentina and Uruguay and traverse Californian coastal habitats and the Great Plains of North America on their way to the breeding grounds in the high Arctic (del Hoyo et al., 1996; Global Register of Migratory Species, http://www.groms.de). Although this mechanism is speculative, the preferential occurrence of the disjunctive South–North American Hordeum species pairs in temporary wet habitats along ditches, riverbanks, and at lake shores certainly promotes epizoochory, where spikelets scattered in mud become attached to the legs of feeding waterbirds (pers. obs.). These habitats are the only sources of open water and feeding grounds for these birds in an otherwise semiarid to arid landscape of the Patagonian steppe. Repeated long-distance dispersal between North and South America is not unusual, as it was also found in the Gentianaceae genus Halenia (von Hagen & Kadereit, 2003), although in the case of Halenia, migration via stepping-stone habitats was proposed.
An explanation for long-distance dispersal from South America to Europe is much harder to find, if one H. secalinum progenitor should indeed belong to the New World clade. Even here bird dispersal seems to be the best explanation. During spring in the northern hemisphere, birds wintering in Patagonia can regularly be found on European coasts, carried away by storms on their way from South to North America (del Hoyo et al., 1996). They might, thus, provide a one-way connection between South America and Europe.
The Mediterranean basin and the Cape Province share similar climatic conditions and several plant families and genera (Raven, 1963). In the light of the dispersal ability of plants (Wagner & Funk, 1995; Vijverberg et al., 1999; Wright et al., 2000; Davis et al., 2002; Fuertes-Aguilar et al., 2002; Coleman et al., 2003; this study), long-distance dispersal is the most parsimonious explanation for the occurrence of Hordeum species at the northern and southern rim of the African continent. These regions are connected by migrating birds such as red knot (Calidris canutus) and corncrake (Crex crex), which both occur in H. secalinum habitats in Europe and the Mediterranean and winter in the Cape Province's habitats of H. capense.