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

This Article Abstract Full Text (PDF) Supplemental Data Submit a response Alert me when this article is cited Alert me when eLetters are posted Alert me if a correction is posted Services Email this article to a friend Similar articles in this journal Similar articles in ISI Web of Science Alert me to new issues of the journal Download to citation manager Citing Articles Citing Articles via HighWire Citing Articles via ISI Web of Science (64) Citing Articles via Google Scholar Google Scholar Articles by Zhang, L.-B. Articles by Kadereit, J. W. Search for Related Content PubMed Articles by Zhang, L.-B. Articles by Kadereit, J. W. Agricola Articles by Zhang, L.-B. Articles by Kadereit, J. W.

Phylogeny and quaternary history of the European montane/alpine endemic Soldanella (Primulaceae) based on ITS and AFLP variation¹

Institut für Spezielle Botanik und Botanischer Garten, Johannes Gutenberg-Universität, 55099 Mainz, Germany

Received for publication January 23, 2001. Accepted for publication June 7, 2001.

	ABSTRACT

TOP ABSTRACT INTRODUCTION MATERIALS AND METHODS RESULTS DISCUSSION LITERATURE CITED

 
Soldanella contains 16 species of herbaceous perennials that are endemic to the central and south European high mountains. The genus is ecogeographically subdivided into forest/montane and alpine species. Evolutionary relationships and large-scale biogeographic patterns were inferred from parsimony analyses of the internal transcribed spacer (ITS) regions of nuclear ribosomal DNA, and genetic distance analyses based on amplified fragment length polymorphism (AFLP) markers. The ITS region proved useful for examining subgeneric relationships and testing hypotheses on genus-wide divergence times, whereas the AFLP markers were suitable for studying relationships among closely related taxa and biogeographic patterns of divergence. Neither ITS nor AFLP data supported sectional delimitations, particularly those related to the grouping of S. alpina (sect. Soldanella) with S. pusilla (sect. Tubiflores), which may be the result of hybridization. Additional results and conclusions drawn are (1) Soldanella is derived from an ancestor of Asian origin with a montane ecology; (2) estimates of divergence times suggest a late Quaternary origin of the genus; (3) alpine species of sect. Tubiflores diverged from within a paraphyletic sect. Soldanella of mainly montane species; (4) alpine and montane species of Soldanella experienced different cycles of range expansion and contraction during late Quaternary climatic changes, resulting in differential patterns of geographic distribution; and (5) AFLP divergence among montane species from eastern Europe was lower than between alpine species; we hypothesize that the latter differentiated in allopatric regions of expansion during glacials, while the former experienced secondary contact at lower elevations in more southern refugia.

Key Words: AFLP • alpine species • historical biogeography • ITS • Primulaceae • Quaternary • Soldanella

	INTRODUCTION

TOP ABSTRACT INTRODUCTION MATERIALS AND METHODS RESULTS DISCUSSION LITERATURE CITED

 
One major goal of modern biogeography is to reconstruct the phylogeny of taxa of similar geographic distribution in relation to the geologic and paleoclimatic history of their distribution area (Cracraft, 1986 ; Riddle, 1996 ). The advent of molecular approaches allows scientists to estimate phylogenies at various taxonomic levels and across a wide range of geographic scales (Avise, 2000) . In recent years, a number of alpine plant species from the Northern Hemisphere have been examined from an intraspecific-biogeographic or "phylogeographic" perspective (reviews by Comes and Kadereit, 1998 ; Stehlik, 2000 ). Most notably, a variety of resultant DNA patterns have been interpreted as resulting from the climatic oscillations of the Quaternary (<1.6 million years [my]; Harland et al., 1990 ), and there has been a resurgence of debate on in situ glacial survival in both the Arctic (Abbott et al., 1995, 2000 ; Gabrielsen et al., 1997 ; Tollefsrud et al., 1998 ; Hagen, Giese, and Brochmann, 2001 ) and the Alps (Stehlik, Schneller, and Bachmann, 2001) . In contrast, few studies on alpine plant taxa have dealt with both molecular systematic relationships at the supraspecific level and a broad coverage of geographic areas (e.g., Hungerer and Kadereit, 1998 ; Conti et al., 1999 ). A set of areas that offers particular promise for such an approach is the central and south European high mountain ranges, essentially the Alps, Pyrenees (and adjacent Cantabrian Mountains), Apennines, Tatra Mountains, Carpathians, and the Balkans (see Fig. 3).

View larger version (36K): [in this window] [in a new window]   Fig. 3. Distribution of the 16 species of Soldanella in the European high mountains based on data from Meyer (1985) and extensive herbarium surveys based on 6300 sheets (L.-B. Zhang and J. W. Kadereit, unpublished data). Symbols indicate narrowly circumscribed regional occurrences. The overall range of S. alpina is shown in the inset. See also Table 1 for the approximate distribution ranges of species and subspecies

During the more recent glaciations of the Quaternary, only the Alps and Pyrenees were extensively covered by massive ice sheets, and locally restricted areas of glaciation occurred in all the other mountain ranges of the European Alpine system (Charlesworth, 1957 ; Messerli, 1967 ). It seems likely that the Quaternary oscillations had a significant impact on the genetic structure of Soldanella, as large areas of its present range were repeatedly covered by ice or affected by local climatic change. One expected result of these cycles, which fragment and isolate habitats and populations, should be the establishment of distinct patterns of molecular divergence. It is also feasible, however, that glacial periods have fostered joint distributional shifts of populations to lower elevations or sympatric refugia, thus either promoting interpopulational gene flow and inhibiting speciation or fostering hybrid speciation after secondary contact of populations diverged in allopatry (Stebbins, 1984 ; Sang, Crawford, and Stuessy, 1995a ; Brochmann, Nilsson, and Gabrielsen, 1996 ). At present, however, we are far from being able to assess for the European Alpine system whether the Quaternary climatic changes have produced geographically (i.e., latitudinally and/or altitudinally) differentiated patterns of molecular divergence or have increased the initiation of species and rates of speciation. The lone exception involves molecular divergence time estimates for Gentiana sect. Ciminalis, which place the origin of this taxon firmly within the Quaternary (Hungerer and Kadereit, 1998 ).

Here, we concentrate on the following questions by examining sequence variation of the ITS regions of nrDNA and patterns of AFLP variation: (1) Can we assess the pattern and timing of how the genus (or its ancestor) reached the European Alpine system? (2) What are the phylogenetic and biogeographic patterns of divergence at the intrageneric level? (3) Does the phylogenetic framework allow us to investigate the order and frequency of changes in habitat preference (alpine vs. montane) among taxa? (4) Can we date the origin of distinct intrageneric lineages associated with alpine/montane habitats? (5) Given that alpine and montane species of Soldanella may have reacted differently to the climatic oscillations of the Quaternary, are there indications of differential patterns of geographic genetic subdivision among these groups?

Finally, results from this study will be compared with the previously studied Gentiana sect. Ciminalis (Hungerer and Kadereit, 1998 ). However, only when spatiotemporal patterns of diversification have been evaluated for a larger number of plant taxa across the European Alpine system can more general statements be made about the importance of Quaternary climatic changes in creating plant diversity in these mountain regions.

	MATERIALS AND METHODS

TOP ABSTRACT INTRODUCTION MATERIALS AND METHODS RESULTS DISCUSSION LITERATURE CITED

 
The study system
Species of Soldanella are easily recognized by their perennial habit, roundish, evergreen leaves, and violet or, rarely, white flowers. These flowers may have either campanulate (sect. Tubiflores) or funnel-shaped (sect. Soldanella) corollas, all with laciniate (incised) lobes. Individuals are primarily insect-pollinated (bees, bumble bees, butterflies) and outcrossing (Lüdi, 1927 ); there is no firm experimental evidence of the self-fertilization in S. alpina suggested by several authors (Kerner von Marilaun and Oliver, 1904 ; Kugler, 1970 ). The majority of species have a reported chromosome number of 2n = 40 (Kress, 1969, 1984b ; Peev, 1977 ), except S. montana and S. villosa (2n = 38) and S. chrysosticta (2n = 38 or 40).

Based on a forthcoming taxonomic treatment (L.-B. Zhang and J. W. Kadereit, unpublished data) the genus is composed of 16 species, three of which contain two or three subspecies. Whereas the two sections are morphologically well defined (von Borbás, 1901 ; Vierhapper, 1904a, 1926 ; Knuth, 1905 ; Honcik, 1963; Pawowska, 1963, 1972 ; Meyer, 1985 ; see Figs. 1 and 2 and Table 2), there are only minute differences among taxa at the intrasectional level, including corolla lobe size, leaf dimensions, and glandular hair structure on scapes and stipes (L.-B. Zhang and J. W. Kadereit, unpublished data). Both species of sect. Tubiflores (S. minima, S. pusilla) grow primarily in high-elevation alpine habitats, where they are largely confined to snow pockets, grassy slopes, and wet meadows above the timberline (Table 1). In contrast, species of sect. Soldanella mainly inhabit montane coniferous and/or deciduous forests and are most common at lower altitudes, at least in the northern part of the genus' range; exceptions are S. alpina and S. rugosa, both of which have an alpine ecology. Most species within the genus have a broad edaphic tolerance, growing on both base-rich and base-poor soils derived from a variety of parent materials (Table 1). Only S. minima is entirely restricted to limestone (or dolomite), whereas several species from the Balkans (S. chrysosticta, S. oreodoxa, S. rhodopaea) and southern Italy (S. calabrella) exclusively grow on acidic (mostly granitic) soils.

View larger version (40K): [in this window] [in a new window]   Figs. 1–2. Morphological characters distinguishing Soldanella (Fig. 1 ) sect. Soldanella (here: S. major) and (Fig. 2 ) sect. Tubiflores (here: S. minima). Section Soldanella is characterized by relatively large (13–35 cm) plants with several flowers and a long style (Fig. 1a ); long corolla lobes and presence of corolla scales (Fig. 1b ); anthers with caudate connective appendages (Fig. 1c ); and ten-toothed capsules (Fig. 1d ). Section Tubiflores has relatively small (<10 cm) plants with only one flower and a short style (Fig. 2a ); short corolla lobes and absence of corolla scales (Fig. 2b ); acuminate connective appendages (Fig. 2c ); and five-toothed capsules (Fig. 2d ). See also Table 2

View this table: [in this window] [in a new window]   Table 2. Morphological characters used to distinguish the two sections of Soldanella according to various authors (von Borbás, 1901; Vierhapper, 1904a, 1926; Knuth, 1905; Honcik, 1963; Pawowska, 1963, 1972; Meyer, 1985). See also Figs. 1 and 2

Plant material
Forty-one and 50 accessions of Soldanella originating from locations throughout the European high mountains were used for the ITS and AFLP analyses, respectively. In both analyses, all 16 species are represented, and for most of the species several intraspecific accessions from different locations were sampled to represent their geographic range.

Previous authors (Diels, 1910 ; Lüdi, 1927 ) and recent cpDNA data from Primulaceae-Primuleae (Anderberg, Ståhl, and Källersjö, 1998 ; Källersjö, Bergqvist, and Anderberg, 2000 ) have suggested close affinities of Soldanella to Primula L. More specifically, in the ndhF/rbcL/atpB phylogeny by Källersjö, Bergqvist, and Anderberg (2000) , Soldanella together with Omphalogramma (Franch.) Franch. constitutes the sister group to a clade comprising Primula and its likely derivatives Cortusa L. and Dodecatheon L. For the ITS analysis, therefore, we included representatives of six sections of Primula as outgroup taxa: P. farinosa L. (sect. Aleuritia), P. latifolia Lapeyr. (sect. Auricula), P. cuneifolia Ledeb. (sect. Cuneifolia), P. tschuktschorum Kjellman (sect. Crystallophlomis), P. parryi A. Gray (sect. Parryi), and P. bulleyana Forrest (sect. Proliferae). Material of Omphalogramma was unavailable. As additional outgroup taxa we included two species of Douglasia Lindley (D. alaskana Coville & Standley ex Hultén, D. beringensis S. Kelso, Jurtsev & D. F. Murray). This genus, together with Androsace L., is sister to the remaining genera of Primulaceae (see above) included in the cpDNA phylogeny by Källersjö, Bergqvist, and Anderberg (2000) . As outgroups in the AFLP analysis we used P. latifolia and P. bulleyana.

Leaf material for DNA analysis was generally obtained (as silica gel-dried material) from natural populations or directly taken from herbarium specimens. For a very few accessions, leaf material was obtained from greenhouse-grown plants raised from seed provided by botanical gardens. The ITS sequence of Primula cuneifolia was kindly provided by Noriyuki Fujii (Tokyo), while those of P. farinosa/P. tsckuktschorum and the two Douglasia species were derived via GenBank (data from Conti et al., 2000 ). Full details on all accessions used for ITS and/or AFLP analysis are listed in the supplemental Appendix (http://ajbsupp.botany.org/v88/zhang.pdf) along with their source and place of origin.

DNA isolation
Total genomic DNA was isolated and cleaned with NucleoSpin-Plant (Macherey-Nagel, Düren, Germany) or DNeasy (Qiagen, Hilden, Germany) extraction kits, using 10 x 10 mm dried (or, rarely, fresh) leaf material. DNA concentrations were determined by spectrophotometry with a GeneQuant RNA/DNA calculator (Pharmacia, Uppsala, Sweden).

PCR amplification of the ITS region
For most accessions the entire ITS-1, 5.8S, and ITS-2 region was polymerase chain reaction (PCR)-amplified using external primers "ITS-A" (5'-GGAAGGAGAAGTCGTAACAAGG-3'; Blattner, 1999 ) and "ITS-4" (5'-TCCTCCGCTTATTGATATGC-3'; White et al., 1990 ). For some herbarium material, it was necessary to amplify the entire ITS region using internal "ITS-C" (5'-GCAATTCACACCAAGTATCGC-3') and "ITS-D" (5'-CTCTCGGCAACGGATATCTCG-3') primers (Blattner, 1999 ) in combination with "ITS-A" and "ITS-4," respectively. All amplifications were performed in 25-µL volumes containing 19.3 µL dionized sterile water, 1.25 µL of 50 mmol/L MgCl₂ solution, 2.5 µL 10x BioTherm buffer (GeneCraft, Münster, Germany), 0.25 µL of a 20 mmol/L dNTP solution in equimolar ratio, 0.25 µL of each primer at 50 pmol/µL, 1 unit (0.2 µL) BioTherm polymerase (GenCraft), and 1 µL genomic DNA (5–100 ng). Double-stranded DNA templates were produced in an Autogene II thermal cycler (Grant Instruments, Cambridge, UK) set for an initial 1 min at 94°C, followed by 35 cycles of 94°C for 0.3 min, 55°C for 0.5 min, and 72°C for 1 min, with two final incubations of 1.3 min at 55°C plus 8 min at 72°C. The PCR products were cleaned with NucleoSpin-Extract (Macherey-Nagel) or QIAquick (Qiagen) purification kits.

DNA sequencing
Purified DNAs were directly sequenced using the BigDye Terminator Cycle Sequencing Kit (Applied Biosystems, Norwalk, Connecticut, USA) with AmpliTaq DNA polymerase. Primers were the same as those of the intitial PCR and used singly in forward and reverse reactions. The thermocycling profile consisted of an initial 1 min at 96°C, followed by 27 cycles of 6 s at 96°C, 12 s at 55°C, and 4 min at 60°C and two final incubations of 6 s at 51.4°C plus 4 min at 60°C. Single-stranded, dye-labeled DNAs were precipitated with 2.0 µL 3 mol/L NaOAc (pH 5.2) and 50 µL 96% ethanol, followed by a washing step in 70% ethanol. Sequences were detected on automated sequencers (ABI 373 and 377).

Sequence alignment and analysis
Sequences of forward and reverse strands were automatically aligned using SEQUENCHER (version 3.0, Gene Codes Corporation, Ann Arbor, Michigan, USA) with manual correction. Boundaries of the two ITS regions were determined by comparison with sequences of Dendroseris (Sang et al., 1994 ) and Gentiana (Hungerer and Kadereit, 1998 ). The alignment was straightforward among the Soldanella taxa and required the introduction of only two insertion/deletion events (indels), each of one base pair (bp) in length. The inclusion of Douglasia and Primula sequences in the Soldanella data matrix necessitated the inference of additional indels to adjust the overall alignment: 40 indels of 1 bp, 4 of 2 bp, 3 of 3 bp, 1 of 4 bp, and 1 of 5 bp. In addition, all Soldanella and Primula accessions analyzed shared an 18-bp deletion relative to Douglasia. The fully aligned data matrix is available upon request, and all sequences (including 5.8S) are deposited in the GenBank database (see supplemental Appendix, http://ajbsupp.botany.org/v88/zhang.pdf, for accession numbers).

Phylogenetic analysis of ITS sequences
Unweighted maximum parsimony analyses were conducted on the entire ITS-1-5.8S-ITS-2 region by a heuristic search strategy using PAUP* (Swofford, 2000) with the following options in effect: MULTREES, tree bisection reconnection (TBR) branch swapping, 100 replications of RANDOM sequence entries, and gaps (indels) treated as missing. In a separate search with the settings above, potentially informative indels were scored as additional characters; indel overlapping required the conversion of indels (37 nucleotide sites) into 33 binary (presence/absence) characters and four multistate characters. To assess node support, bootstrap values (BV) were estimated from 100 replicates of full heuristic searches using 100 RANDOM ADDITION sequences of accessions for each replication and with the settings above in effect. PAUP* was also used to calculate the proportion of nucleotide sequence differences (adjusted for missing data and gaps), and Kimura (1980) two-parameter (K2P) distances as the measure of sequence divergence (see below).

Rate-constancy of ITS sequence evolution was evaluated by relative rate tests (Wu and Li, 1985 ) as implemented by the program K2WuLi (Jermiin, 1996 ) and using Douglasia alaskana as outgroup. After discovery of rate constancy among lineages of the ingroup (Soldanella/Primula), approximate ITS sequence divergence rates were extrapolated from the fossil record of Primulaceae-Primuleae. This involved calibrating the molecular clock with the help of an Androsace fossil seed from Siberia, broadly dated to the Miocene (Dorofeev, 1963 ), i.e., 23.3–5.2 mya (Harland et al., 1990 ). Because Douglasia is apparently the closest relative of Androsace (Källersjö, Bergqvist, and Anderberg, 2000) , and is often regarded as a section of the latter (Wendelbo, 1961a ; Kress, 1984a ), we could assign respective maximum and minimum ages to the basalmost node of our ITS tree. Times of divergence for Soldanella were then inferred from the internally calibrated ITS divergence rate. As the measure of differences between sequences, we employed pairwise K2P distances to compensate for multiple substitutions and different substitution rates of transitions and transversions. Ninety-five percent confidence intervals for sequence divergence estimates were approximated as ±2 SE of the K2P distances. Multiple intraspecific accessions with identical sequences were excluded from all sequence divergence calculations and relative rate tests, with single accessions chosen at random to represent each species (asterisked in Fig. 4).

View larger version (42K): [in this window] [in a new window]   Fig. 4. One of 12 equally parsimonious trees of 41 accessions from 16 species of Soldanella based on ITS sequence variation (ITS-1-5.8S-ITS-2), and rooted by Primula and Douglasia species. Gaps (indels) were treated as missing data. Numbers of nucleotide substitutions are shown above branches, and bootstrap values (>50%) from 100 replicates are shown below branches. Dashed lines indicate branches that collapse in the strict consensus. The bar designates a single deletion (1 bp) independently supporting the S. alpina + S. pusilla subclade (see text). Numbers following accessions refer to populations identified in the supplemental Appendix (http://ajbsupp.botany.org/v88/zhang.pdf). Asterisked accessions were used for sequence divergence calculations and relative rate tests. Boldface taxa share an alpine ecology

AFLP data analysis
Raw fragment data were analyzed using GENESCAN (version 3.1, ABI) with the following peak amplitude thresholds in effect: BLUE (60), GREEN (30), YELLOW (40), and RED (40). Presence or absence of peaks (i.e., fragments) was scored for each accession within a readable range of 75–500 bp and assembled as a binary (1/0) matrix using GENOTYPER (version 2.1, ABI). The following GENOTYPER options were chosen: CALCULATE SCALE FACTOR (50–500 bp), and FILTER LABELS (0.0–0.9). Electrophoretograms generated by GENOTYPER were carefully checked individually in order to avoid possible misinterpretations due to automated fragment scoring. In a few cases, fragments were scored as missing data (0.02%) since character states could not be determined unambiguously.

Pairwise genetic distances (D) for all accessions were calculated from the presence/absence matrix according to the complementary value of Nei and Li's (1979) similarity coefficient implemented in PAUP*: D = 1 – SC = 1 – [2n_xy/(n_x + n_y)], where SC is the similarity coefficient, n_xy is the number of identical fragments shared between two accessions, and n_x and n_y are the total number of fragments in accessions x and y, respectively. A neighbor-joining (NJ) phenogram (Saitou and Nei, 1987 ) was then calculated based on the distance matrix and 1000 bootstrap replicate data sets using PAUP*. To further illustrate interspecies relationships, principal coordinates (PCOs) were extracted from a matrix of squared Euclidean distances between all pairs of accessions and projected in three dimensions (programs DCENTER, EIGEN, and MOD3D of NTSYS-pc, version 2.0; Rohlf, 1997 ). Relative locations in space of the AFLP phenotypes were inferred from a minimum-length spanning tree (MST) computed from the latter distance matrix and superimposed onto the PCO plot (Rohlf, 1997 ).

	RESULTS

TOP ABSTRACT INTRODUCTION MATERIALS AND METHODS RESULTS DISCUSSION LITERATURE CITED

 
Characteristics of the ITS region in Soldanella
Characteristics of the ITS-1, 5.8S, and ITS-2 sequences of Soldanella are summarized in Table 3. The ITS-1 region (249 bp) was only slightly longer than ITS-2 (230–231 bp). The 5.8S gene was uniformly 165 bp in length. Of the 645 aligned positions, only two sites involved gaps in ITS-2. When accessions of Douglasia and Primula (outgroups) were taken into account, there were 679 aligned positions, and the total number of gaps involved increased to 50 (ITS-1: 30; ITS-2: 15; 5.8S: 5). This clearly reflected the distinct sequences of Douglasia/Primula and Soldanella (see below). Within Soldanella, the number of sites that varied in more than one taxon, and thus were potentially parsimony-informative, were only 7 and 8 (of 30 variable sites) in ITS-1 and ITS-2, respectively. With the outgroup taxa included, 140 sites (of 172 variable sites) were potentially informative in ITS-1, 91 (of 114 variable sites) in ITS-2, and 6 (of 8 variable sites) were still recorded in the 5.8S gene (Table 3).

Phylogenetic analysis of nrDNA
Unweighted parsimony analysis of the 49 accession/679 character matrix (including outgroups), with gaps scored as missing data, generated 12 equally parsimonious trees of 512 steps with a consistency index of 0.816 (excluding uninformative characters) and a retention index of 0.878. Essentially identical tree topologies emerged when gaps were recoded as separate characters (not shown), although this improved the support of a single clade within Soldanella (see below). One of the shortest trees (without gap treatment), which is almost identical to the strict consensus, is shown in Fig. 4.

The monophyly of Soldanella (BV = 100%) was clearly supported, and the 16 species (represented by 41 accessions) were basally resolved into two clades, A and B, albeit with only low to moderate support. These groups, however, did not correspond to the two sections of Soldanella. Species of sect. Tubiflores (S. minima, S. pusilla), together with two species traditionally assigned to sect. Soldanella (S. alpina, S. villosa), were all members of the same moderately supported monophyletic group (clade A; BV = 62%). Two subclades appeared within this group, one containing all accessions of S. minima (BV = 89%) and the other uniting all samples of S. alpina and S. pusilla (BV = 67%). The alpina-pusilla subclade was also supported by a 1-bp deletion, recoding of which as a separate character significantly improved the bootstrap value for this subclade (89%). In general, clade A comprised species that are mainly distributed in the alpine regions of the western and central parts of the European high mountains (i.e., Pyrenees, Alps, Apennines). Notable exceptions are S. villosa, restricted to montane habitats in northern Spain (Cantabria/western Pyrenees), and three accessions of the alpine S. pusilla from disjunct populations in the South Carpathians (Bucegi and Retezat Mountains) and the southern Balkans (Pirin Mountains). For convenience, clade A will be referred to as the "western-central group."

In contrast, clade B (BV = 54%) was largely composed of montane species of sect. Soldanella that are distributed in the eastern part of the European Alpine system. Within this "eastern group," there were two groups that received further support as monophyletic subclades. In particular, accessions of S. chrysosticta, S. hungarica, S. major, S. pindicola, and S. rhodopaea formed a strongly supported subclade (BV = 95%). With the exception of accessions of S. major from the eastern Alps this subclade only contained samples from the southern Balkans and the southern Carpathians. A second subgroup that could be distinguished included S. angusta, S. carpatica, S. marmarossiensis, S. montana, S. oreodoxa, and S. rugosa (BV = 62%). This subclade comprised accessions that originate from the eastern Alps/Bohemian Massif, the Tatras, and the north-central Carpathians. The rather unexpected position of S. calabrella from Calabria (southern Italy) as sister to this latter subclade received moderate support (BV = 60%).

AFLP analysis
In the 50 Soldanella accessions of the 16 species studied, the three primer combinations amplified a total of 690 fragments, 683 of which were polymorphic (Table 4). Of those, 511 were shared by two or more accessions, and thus were potentially informative, whereas 172 were either unique to a single accession or present in all except one. When the two outgroup accessions of Primula were included, the total number of fragments increased to 731. The selective primer pair combinations varied in their ability to detect AFLP variation across Soldanella accessions, with numbers of detected polymorphic fragments ranging from 182 (E40/M54) to 281 (E39/M61). The percentage of potentially informative fragments (out of the total) generated by each selective primer combination varied between 68.5% (E45/M49) and 81.3% (E39/M61).

View larger version (43K): [in this window] [in a new window]   Fig. 5. Neighbor-joining phenogram of 50 accessions from 16 species of Soldanella and two outgroup taxa of Primula based on Nei and Li (1979) genetic distances for 731 polymorphic AFLP fragments obtained with three primer combinations (see Table 4 ). Bootstrap values (>50%) based on 1000 replicates are shown along branches. Accession numbers refer to populations given in the supplemental Appendix (http://ajbsupp.botany.org/v88/zhang.pdf). Scale bar indicates a genetic distance of 0.01

Figure 6 illustrates the multidimensional relationships among the 50 AFLP phenotypes of Soldanella based on their Euclidean distances. The first three principal coordinates accounted for 21.8%, 10.1%, and 7.5% of the total variance. Accessions of the montane species from the east (corresponding to cluster B and thus including the alpine S. rugosa) were clearly separated from the remaining material by the first coordinate, thus confirming the first division in both the ITS and AFLP trees. In conjunction with the MST analysis, combinations of the second and third coordinate further subdivided the remaining material into four groups: (1) S. alpina, (2) S. minima, (3) S. pusilla, and (4) the single accession of S. villosa. Somewhat unexpectedly, the MST analysis identified nearest-neighbor links between S. alpina and various other accessions (S. minima, S. villosa, and the montane group) rather than S. pusilla, as was suggested by both the ITS and AFLP tree topologies. Overall, this multidimensional approach fully established a greater separation among the alpine species (S. alpina, S. minima, S. pusilla) than among the montane species from the east.

View larger version (17K): [in this window] [in a new window]   Fig. 6. Principal coordinates analysis of 683 polymorphic AFLP characters for 50 accessions of Soldanella based on their squared Euclidean distances. A minimum-length spanning tree (Rohlf, 1997 ) is superimposed onto the plot

	DISCUSSION

TOP ABSTRACT INTRODUCTION MATERIALS AND METHODS RESULTS DISCUSSION LITERATURE CITED

 
Extrageneric relationships and monophyly of Soldanella
Källersjö, Bergqvist, and Anderberg (2000) recently presented a cpDNA analysis of Primulaceae-Primuleae and showed that (1) Soldanella, together with Omphalogramma, constitutes the sister group to a paraphyletic Primula containing Cortusa and Dodecatheon and (2) these two groups together are sister to Androsace and Douglasia (see MATERIALS AND METHODS). Of those genera of Primulaceae-Primuleae not included by Källersjö, Bergqvist, and Anderberg (2000) , Pomatosace Maxim. and Vitaliana Sesler have been considered to be closely related to Androsace (Pax, 1905 ; Constance, 1938 ; Anderberg and Ståhl, 1995 ), Dionysia Fenzl and Hottonia L. to be closely related to Primula (Wendelbo, 1961b ; Anderberg and Ståhl, 1995 ), and the monotypic Bryocarpum Hook. f. & Thoms. to be either sister to or part of Omphalogramma (Fletcher, 1949 ; Richards, 1993 ; Anderberg and Ståhl, 1995 ). If these hypotheses about generic relationships within Primulaceae-Primuleae are correct, Soldanella would form a well-established clade together with Bryocarpum and Omphalogramma. Unfortunately, material of these latter two genera was not available for the present study. We feel, however, that the use of Primula and Douglasia as outgroup taxa for our analysis of Soldanella is justified because they represent the next two neighboring clades of Soldanella in the phylogeny proposed by Källersjö, Bergqvist, and Anderberg (2000) . Due to the unavailability of the Asian Bryocarpum and Omphalogramma, it was not possible to test the monophyly of Soldanella. Such monophyly, however, appears highly likely because Soldanella is an isolated European endemic and has a unique flower morphology with five laciniate corolla lobes that is otherwise unknown in Primulaceae-Primuleae (Pax, 1905 ; Hu and Kelso, 1996 ).

Sectional subdivision of Soldanella
Probably the most unexpected and important phylogenetic aspect of our analysis of Soldanella is the observation that both the ITS and AFLP data do not confirm the traditional and morphologically well-founded subdivision of Soldanella into two sections (Table 2). Instead, the ITS-derived phylogeny of Soldanella shows that members of these sections are found in two distinct clades, A and B (Fig. 4). Thus, S. alpina (sect. Soldanella) and S. pusilla (sect. Tubiflores) form a moderate to well-supported subclade (depending on gap treatment), which together with S. villosa (sect. Soldanella) and S. minima (sect. Tubiflores) forms a moderately supported monophyletic clade (A); this, in turn, is sister to a weakly supported clade (B) of all the remaining taxa of sect. Soldanella. However, given the generally low bootstrap support for various internal clades, it is possible to argue that our ITS phylogeny essentially represents an unresolved polytomy of at least four lineages, i.e., S. villosa, S. minima (89%), S. pusilla + S. alpina (67 or 89%), and clade B (54%). Even such a polytomy cannot be resolved into two monophyletic sections because of the moderate to strong support for the subclade uniting S. pusilla (sect. Tubiflores) and S. alpina (sect. Soldanella) and the concomitant lack of support for a monophyletic sect. Tubiflores (S. minima, S. pusilla).

The AFLP phenogram (Fig. 5) is remarkably congruent with the ITS phylogeny in overall topology in that it shows the existence of two main lineages (clusters A and B) and a greater genetic similarity of S. alpina to S. pusilla than to the remainder of sect. Soldanella. The AFLP data, however, clearly place S. villosa as sister to the remaining species, rather than within cluster A as shown by the ITS tree. Taken at face value, the AFLP topology is entirely consistent with the hypothesis that sect. Soldanella is paraphyletic in relation to a nonmonophyletic sect. Tubiflores, and such interpretation is not in conflict with the ITS phylogeny (Fig. 4).

While the resolution of a paraphyletic sect. Soldanella, with a basal position of S. villosa, seems plausible from a biogeographic perspective (see below), the nonmonophyly of sect. Tubiflores is more difficult to understand. In comparison to Omphalogramma/Bryocarpum, the only unambiguous synapomorphy defining this section is a campanulate corolla shape (L.-B. Zhang and J. W. Kadereit, unpublished data). Consequently, if sect. Tubiflores is assumed to be diphyletic, this character would have arisen in parallel. However, separate analyses (not shown) that excluded the problematic S. alpina from both the ITS and AFLP data sets resulted in trees that recovered sect. Tubiflores as a monophyletic group, although with insufficient bootstrap support (<50% in both the ITS and AFLP trees). Regardless of such speculation, it would not change the proposed conclusion that the alpine species of sect. Tubiflores (S. minima, S. pusilla) arose from within a paraphyletic sect. Soldanella.

Also contributing to the nonmonophyly of sect. Tubiflores is the unexpected sister group relationship between S. pusilla (sect. Tubiflores) and S. alpina (sect. Soldanella), as most strongly supported by the ITS data (Fig. 4). In fact, the ITS sequences of S. pusilla and S. alpina are identical with the exception of a single base substitution each in two intraspecific accessions of S. alpina. In contrast, all other species of sect. Soldanella are distinguished from members of sect. Tubiflores by at least five base substitutions. Barring lineage sorting, such pattern of discordance between the ITS phylogeny and taxonomic classification of S. alpina can be interpreted to imply phenotypic convergence or hybridization. However, the possibility of convergent evolution appears highly unlikely unless the entire complex of characters distinguishing sect. Tubiflores and sect. Soldanella (see Figs. 1 and 2; Table 2) are regulated by one or a few major genes with strongly pleiotropic effects. If interpreted to imply hybridization, as will be argued here, the near identity in ITS sequences between S. pusilla and S. alpina cannot be explained by recent gene flow. Although our sampling included various accessions of S. alpina from outside its distributional overlap with S. pusilla (i.e., Cantabria, Pyrenees), none of these accessions were positioned next to members of sect. Soldanella (Fig. 4). Instead, our data suggest that S. alpina (2n = 40) is a likely hybrid derivative of S. pusilla (2n = 40) and an unknown member of sect. Soldanella (of identical chromosome number). Such a hypothesis of homoploid hybrid speciation (Arnold, 1997 ; Rieseberg and Carney, 1998 ) would necessitate that concerted evolution has occurred in individuals and populations of S. alpina and was fast enough to homogenize ITS repeats of sect. Soldanella in the direction of S. pusilla (see also Sang, Crawford, and Stuessy, 1995a ; Wendel, Schnabel, and Seelanan, 1995a ).

The AFLP phenogram also clustered S. alpina with S. pusilla, although with insufficient support (Fig. 5). When the distribution of individual AFLP fragments across lineages of Soldanella was reexamined by using MacClade (version 3.07; Maddison and Maddison, 1992 ), 50 out of 511 potentially informative fragments were found to be exclusively shared by S. alpina and a varying number of other taxa of sect. Soldanella and thus were absent from S. minima and S. pusilla. In contrast, 27 fragments were entirely restricted to S. alpina/S. minima and/or S. alpina/S. pusilla. This pattern of fragment sharing can be interpreted as a molecular signature for the involvement of a member of sect. Soldanella in the origin of S. alpina. A subsequent survey of cpDNA (rpl16 intron) sequences of all the species of sect. Tubiflores and representatives of sect. Soldanella (S. hungarica, S. marmarossiensis, S. montana, S. major) failed to provide further insights (not shown) although this analysis identified the presence of a 2-bp indel shared by S. alpina, S. montana, and S. minima (L.-B. Zhang, unpublished data).

There is additional, nonmolecular evidence in favor of a hybrid origin of S. alpina. First, the hybrid between S. pusilla and S. alpina (S. xhybrida) is reported frequently in areas of distributional overlap throughout the Alps (Huter, 1873 ; Kress, 1984b ), and, based on extensive herbarium surveys (L.-B. Zhang and J. W. Kadereit, unpublished data), is the most frequently observed hybrid within the genus. Although this could simply reflect the large distributional overlap between S. pusilla and S. alpina, the involvement of S. pusilla in the origin of S. alpina provides an alternative hypothesis for this biased pattern of hybrid formation. Second, it has been repeatedly noted (Huter, 1873 ; Kerner von Marilaun, 1875 ; Vierhapper, 1904a, b ) that intersectional hybrids of Soldanella usually resemble the sect. Soldanella parent. This dominance in morphological characters would explain the assignment of S. alpina to sect. Soldanella, despite its putative hybrid nature. Third, S. alpina apparently combines the morphology of sect. Soldanella with the ecology of sect. Tubiflores in terms of habitat preference for high-altitudinal sites. Finally, we note that the exceptionally broad distributional range of S. alpina (Fig. 3, inset) may have resulted from hybridization facilitating range expansion into novel environments not occupied by either parent. Such an ability of hybridization (and introgression) to increase the invasive nature of plant species is well documented (Ellstrand and Schierenbeck, 2000) .

In summary, we conclude that a mono- or diphyletic sect. Tubiflores was derived from a paraphyletic sect. Soldanella and that S. alpina might be considered a hybrid taxon between S. pusilla and an unknown member of sect. Soldanella.

Sequence divergence rates in Primulaceae-Primuleae
Based on an Androsace fossil seed from the Miocene (Dorofeev, 1963 ), we assumed an early (23.3 mya) or late (5.2 mya) boundary for the time of divergence between Androsace/Douglasia and the remainder of Primulaceae-Primuleae, including Soldanella. Accordingly, our approximate estimates of the time at which Soldanella began to diverge, 0.62 vs. 0.14 mya, depend on what ITS substitution rate is used, i.e., 1.67 or 7.47% per my, respectively. This latter rate is exceedingly fast in comparison with previously published rates of ITS divergence (per my) in other plant taxa (Dendroseris: 0.79–1.21%, Sang et al. [1994 ]; Robinsonia: 1.57%, Sang, Crawford, and Stuessy [1995b ]; Gossypium: 0.54–1.0%, Wendel, Schnabel, and Seelanan [1995b ]; Gentiana: 3.2–4.4%, Hungerer and Kadereit [1998 ]; Eupatorium: 1–3%, Schmidt and Schilling [2000 ], and further references therein), and for this reason is less likely. Whichever rate may be correct, even the slow rate of 1.67% per my places the origin of Soldanella firmly within the Quaternary.

Biogeography of Soldanella
On the basis of the current molecular and taxonomic information, Soldanella has its closest relatives in Omphalogramma and Bryocarpum, whose distributions are centered in the Himalayas and southwest China (Pax, 1905 ; Hu and Kelso, 1996 ). Thus, the genus might belong to that group of plant taxa which entered the European Alpine system from central Asia (e.g., Primula, Gentiana, Pedicularis, Saussurea; Ozenda, 1988 ). The discovery that the montane Soldanella villosa is the possible sister taxon to the remainder of the genus according to the AFLP data suggests that the ancestor of Soldanella (or the common ancestor of Soldanella and Omphalogramma/Bryocarpum) was a montane taxon. This agrees with the observation that Bryocarpum and several species of Omphalogramma, irrespective of their distribution at 2200–4000 (4700) m above sea level (asl) altitude in the Himalayas, have a montane ecology with woodland, shrub, and Pinus forest habitats (Pax, 1905 ; Hu and Kelso, 1996 ).

Early history
Despite the vagaries of our molecular clock, Soldanella appears to have originated within the Quaternary and thus more recently than previously hypothesized for other European high mountain plant taxa of central Asian stock (e.g., Primula, Gentiana; Jerosch, 1903 ; Ozenda, 1988 ). By late Tertiary (Pretegelen)/early Quaternary times, mean annual temperatures in Europe had dropped substantially from a mid-Tertiary (Eocene) value of 20°C to 8°–10°C (Kahlke, 1994 ), but the climate across northern Eurasia was still mild and humid enough to support a contiguous belt of mixed deciduous/coniferous forests north of 50°N (Frenzel, 1968 ). Such a potential dispersal corridor for forest herbs, which in later parts of the Quaternary became increasingly blocked by more arid (forest-) steppe and tundra vegetation, may have allowed the westward migration of the montane-Asian ancestor of Soldanella, perhaps during one of the early interglacials (Tegelen, Waal) of the early Quaternary (1.6–0.9 mya).

The distribution of S. villosa, which is the likely sister taxon to the remaining species based on the AFLP data, in the very west of the generic range may indicate that much of the present range of the genus was already occupied in the early history of the genus. In addition, the properties of the low altitude (70-600 m asl) sites of S. villosa in Cantabria and the western Pyrenees, with mostly deciduous forests (Fraxinus, Quercus, Alnus) and mild-humid climates (Jovet, 1970 ), may offer important insights into the ecological requirements of the early Quaternary ancestor of Soldanella. In this context, it is of interest that the distribution of S. villosa in Cantabria and the western Pyrenees closely parallels the natural range of the late Tertiary relic Meconopsis cambrica (L.) Vig., the only European representative of this otherwise Himalayan genus (Jork and Kadereit, 1995 ; Kadereit, Schwarzbach, and Jork, 1997 ). Also, Saxifraga conifera Coss. & Durieu, which recently has been identified as the most basal species within the south European/north African sect. Saxifraga (Vargas, 2000) , is also an endemic of Cantabria and the western Pyrenees. Overall then, these examples may testify to the relictual properties of these mountain ranges.

The origin of the alpine taxa
Following the establishment of Soldanella in Europe, the major phylogenetic and biogeographical event in the history of the genus was the origin of the high-altitude species S. minima and S. pusilla (sect. Tubiflores). Unfortunately, we were not able to date the origin of these species due to low levels of ITS sequence divergence and uncertainties surrounding their phylogenetic (i.e., mono- or diphyletic) status. However, based on the AFLP data, it is clear that these alpine species have originated more recently than the montane S. villosa, and thus probably within the later part of the Quaternary. A similar reasoning applies to the alpine S. rugosa from the East Carpathians, which clearly is derived from within the monophyletic group of montane species of sect. Soldanella from eastern Europe and the eastern Alps (Fig. 5). Thus, two shifts to alpine habitats have certainly occurred within Soldanella.

The origin of alpine plant taxa from lowland forms has long been hypothesized for taxa of the European Alpine system (Christ, 1867) , but has generally been associated with pre-Quaternary events (e.g., Braun-Blanquet, 1923 ; Ozenda, 1988 ; but see Gams [1933] for an early opposing view). Although such timing may be correct for other alpine plant taxa, it is not for Soldanella. To our knowledge, this is the first time that the origin of European alpine plant taxa from montane taxa has been demonstrated by using molecular data and that this origin has been dated (albeit roughly) by using a molecular clock.

The montane clade of sect. Soldanella
Patterns of species and range formation are difficult to discern within the major clade of mainly montane species of sect. Soldanella because relationships are often not well resolved. However, it is evident from the AFLP topology (Fig. 5) that there is little geographical overlap between taxon clusters, and geographically adjacent taxa generally cluster together, with the exception of S. calabrella/S. carpatica. Moreover, southern and probably older species are genetically more differentiated in the AFLP tree when compared to northern species, which seem to have diverged more recently. The biogeographic signature within this AFLP data set is suggestive of a fairly ordered series of vicariance and/or sequential colonization events rather than rampant dispersal. That said, one intraspecific and probably recent dispersal event may involve S. major, which shows a marked range disjunction between the southern Carpathians and the eastern Alps but no indication of incipient "phylogeographic" differentiation among these disjunct populations in terms of AFLP variation.

The striking species richness and distinct biogeographic structure of Soldanella in eastern Europe is not too surprising. Based on fossil pollen evidence, this region served as a major glacial refuge for deciduous and coniferous forests of temperate Europe (e.g., Abies, Alnus, Fagus, Picea; Huntley and Birks, 1983 ); this has also been recognized by recent molecular studies (for reviews see Comes and Kadereit, 1998 ; Taberlet et al., 1998 ). Thus, during the climatic oscillations of the Quaternary, the Balkan mountains may have provided climatically rather stable habitats for montane species of Soldanella. Clearly, the variable relief and habitats of these mountain ranges is conducive to both survival and speciation processes by allowing species to persist by altitudinal shifts and to diverge because of distributional subdivision (Hewitt, 1996, 2000 ).

Quaternary history of montane vs. alpine species
Species of different altitudinal distribution and habitat preference, as found in Soldanella, must have reacted differently to the climatic oscillations of the Quaternary in terms of changes in geographic range, both latitudinally and altitudinally. It seems likely that montane forest habitats were considerably reduced in the northern range of the genus during glacials, whereas the cooler climate allowed alpine meadows and grassland to expand southward. As a consequence, species of Soldanella associated with these habitats must have experienced a different cycle of range expansion and contraction into refugia. Two alpine species, S. minima and S. pusilla, may provide prime illustrations of latitudinal range expansions during glacials. Today, these species are largely restricted to the Alps (i.e., their interglacial refuge), but the presence of disjunct subspecies of these two taxa in central Italy and the southern Balkans, respectively, may reflect their formerly wider distribution in the south during glacial periods (see Fig. 3). On the other hand, the AFLP topology (Fig. 5) of the montane species from the east reveals an obvious decline in genetic distance among species from south to north. It is feasible that this group colonized and diverged from the south in recent interglacial and/or early postglacial times. When a warming climate caused forest habitats of coniferous/deciduous trees (e.g., Abies, Picea, Fagus) to move upslope (e.g., Lang, 1994; Bozilova and Tonkov, 2000 ), associated southern and possibly older taxa, such as S. pindicola, S. rhodopaea, and S. chrysosticta, were isolated first and left behind on the mountain tops of the southern Balkan Peninsula (i.e., their glacial refuge).

Overall, there was less genetic divergence among the montane species of Soldanella from the east than among the alpine species (S. alpina, S. pusilla, S. minima) based both on the AFLP phenogram and the PCO procedures (Figs. 4 and 5). Given that montane forest habitats were lowered during glacial periods (e.g., below altitudes of 1000 m asl in the Bulgarian Rila Mountains; Bozilova and Tonkov, 2000) , low levels of genetic divergence among associated species of Soldanella may result from frequent opportunities for secondary contact at low elevations during glacials. Notably, such glacial intervals of the late Quaternary (<0.9 mya) are thought to have lasted for 100 000 yr, whereas interglacial periods only lasted for 41 000 yr (Bennett, 1997 ). This raises the intriguing possibility that, for most of the time, these montane species have been restricted to low elevations at southern latitudes, which then set the stage for intergradation. Although it is tempting to speculate about a single (or sympatric) glacial refuge in the southern Balkans over extended time periods, there was an unexpected phylogeographic pattern within the group of northern and possibly younger montane species. In fact, both AFLP and ITS topologies (Figs. 3 and 4) revealed a genetic divide between north/central and south Carpathian accessions, which cannot be explained by physical conditions alone. Rather, this pattern may reflect isolation and survival of at least the north/central Carpathian species in a separate glacial refuge. In fact, during the last ice age (Würm/Weichselian) there were glaciers in many parts of the Carpathian Arc, particularly in the south (e.g., Retezat and Bucegi Mountains; Charlesworth, 1957 ). These glaciers may have acted as barriers to, and thus "trapped," the southward retreat of the north/central species, while the southern ones could retire further south.

As regards the larger amount of divergence observed among the alpine taxa, this may at first seem puzzling, since their range expansion over relatively long glacial intervals should have provided multiple opportunities for intergradation. This paradox, however, is probably best explained in terms of geographic displacement into allopatric regions of expansion. Evidence in support of such a scenario is provided, again, by S. minima and S. pusilla, with possibly different trajectories of range expansion to the peninsulas of Italy and the Balkans, respectively. Moreover, S. minima is entirely restricted to limestone (or dolomite) rocks, whereas S. pusilla is tolerant of various soil substrates (Table 1), though with a preference for granitic rocks (L.-B. Zhang and J. W. Kadereit, unpublished data). It appears likely, therefore, that different routes of expansion have tracked differences in substrate specificity. Other events that may have promoted the isolation of alpine species of Soldanella during glacials include the direct dissection of species' ranges by glaciers, or differential in situ survival on mountain tops above the ice sheet ("nunataks").

We emphasize that the different patterns of genetic divergence observed among alpine and montane groups of Soldanella species could be caused by factors other than differential isolation during glacials. For example, lower levels of reproductive isolation within the group of montane species from the east may have decreased their level of interspecific genetic subdivision. However, because we see separate groupings in the AFLP tree of these taxa (Fig. 5), there is no reason to suspect ongoing interspecific gene flow among them. Clearly, more detailed experimental and population level/phylogeographic studies are necessary to illuminate the mechanisms maintaining their integrity, despite their geographic proximity and morphological similarity (L.-B. Zhang and J. W. Kadereit, unpublished data). Another possible factor by which differences in genetic subdivision could have arisen relates to differing levels of inbreeding. However, there is no detailed information available on the breeding system in Soldanella.

Comparison with Gentiana sect. Ciminalis and conclusions
In comparison to Gentiana sect. Ciminalis, the only supraspecific European high mountain endemic for which a molecular-biogeographic scenario has been formulated (Hungerer and Kadereit, 1998 ), no similarities in overall biogeographic pattern can be found. In fact, the Soldanella pattern, with its characteristic west/east embrace of the European Alpine system by S. villosa and the remainder of sect. Soldanella (excluding the putative hybrid S. alpina), strongly contrasts with a more or less linear differentiation from east to west as reported for Gentiana sect. Ciminalis. On the other hand, the age of this latter section has been estimated to be 0.6–1.88 my (Comes and Kadereit, 1998 ; Hungerer and Kadereit, 1998 ), and thus includes the upperbound divergence time (0.62 my) estimated for Soldanella. Such overall pattern of temporal concordance and biogeographic discordance between the two taxa might be attributed to differences in their ancestor's ecological requirements and their