• Primula sect. Aleuritia subsect. Aleuritia (Aleuritia) includes diploid, self-incompatible heterostyles and polyploid, self-compatible homostyles, the latter generally occurring at higher latitudes than the former. This study develops a phylogenetic hypothesis for Aleuritia to elucidate the interactions between Pleistocene glacial cycles, biogeographic patterns, ploidy levels and breeding systems.
• Sequences from five chloroplast DNA loci were analyzed with parsimony to reconstruct a phylogeny, haplotype network, and ancestral states for ploidy levels and breeding systems.
• The results supported the monophyly of Aleuritia and four major biogeographic lineages: an amphi-Pacific, a South American, an amphi-Atlantic and a European/North American lineage. At least four independent switches to homostyly and five to polyploidy were inferred.
• An Asian ancestor probably gave origin to an amphi-Pacific clade and to a lineage that diversified on the European and American continents. Switches to homostyly occurred exclusively in polyploid lineages, which mainly occupy previously glaciated areas. The higher success of the autogamous polyploid species at recolonizing habitats freed by glacial retreat might be explained in terms of selection for reproductive assurance.
The evolutionary history of arctic-alpine plants has been deeply influenced by the repeated cooling episodes of the Pleistocene (Dynesius & Jansson, 2000; Abbott & Brochmann, 2003; Hewitt, 2004). A large proportion of arctic–alpine plants are polyploids supposedly derived from hybridization between diploid progenitors, followed by chromosome doubling (i.e. allopolyploidy) (Stebbins, 1950; Löve & Löve, 1975; Brochmann et al., 2004). Many polyploids appear to rely on autogamous fertilization, while their proposed diploid progenitors are often allogamous, suggesting a causal link between polyploidy and autogamy (Stebbins, 1950; Thompson & Lumaret, 1992). The ‘secondary contact model’, first proposed by Stebbins (1984, 1985) without explicit reference to breeding systems, may be extended to explain the frequent association between polyploidy, autogamy and current patterns of distribution in arctic-alpine plants. According to this model, glacial advancement during the Pleistocene caused the fragmentation of diploid, allogamous populations, which may have survived in ice-free areas (i.e. refugia). As glaciers retreated, the differentiated diploid populations came into contact again and hybridized, giving origin to polyploid, autogamous taxa. Unreliability of pollinators and/or shifts in pollinator faunas associated with Pleistocene climate cycles might have provided the selective forces that favored the establishment of the newly formed polyploid, autogamous species and their higher ability to recolonize deglaciated areas, as compared with their diploid, allogamous progenitors (Stebbins, 1957).
Extensive polyploidization attributed to cyclic population fragmentation and expansion during the Pleistocene has been documented in several genera, including Cerastium, Draba, Parnassia, Saxifraga and Vaccinium (reviewed in Abbott & Brochmann, 2003; Brochmann et al., 2004). However, only a few studies have tried to infer the phylogenetic and biogeographic relationships of arctic-alpine polyploid complexes on a global scale. Examples include Draba and Cerastium (Grundt et al., 2004; Scheen et al., 2004). Furthermore, the combined effects of glacial cycles on ploidy levels and breeding systems have never, to our knowledge, been examined in light of an explicit phylogenetic framework.
Given this background, the study of widespread arctic-alpine polyploid complexes is of special importance in trying to elucidate the interactions between Pleistocene cyclic climate changes, patterns of biotic distribution, polyploidization, and reproductive biology. Primula sect. Aleuritia subsect. Aleuritia Duby (Primulaceae Vent., – hereafter called Aleuritia) represents an ideal case study to investigate such interactions because it is a circum-boreal group that displays variation of ploidy levels, ranging from diploidy to 14-ploidy, and breeding systems, alternating between heterostyly and homostyly.
Heterostyly in Primula is characterized by two floral morphs (distyly) that differ in the reciprocal positioning of stigma and anthers, usually coupled with a self-incompatibility system that prevents self and intramorph fertilization (Wedderburn & Richards, 1992). Conversely, homostyles are self-compatible (Arnold & Richards, 1998) and may derive from heterostylous progenitors via recombination within the heterostyly linkage group, known as the S supergene (Dowrick, 1956; Wedderburn & Richards, 1992).
Aleuritia, recently supported as monophyletic in genus-wide chloroplast-based phylogenies (Mast et al., 2001, 2006), includes 21 arctic-alpine species of small to medium-sized plants, usually perennial, that typically have a single umbel of flowers, a yellow annulus at the center of the corolla, presence of dense farina on the stem and calyx, syncolpate pollen, and a base chromosome number of x = 9 (Richards, 2002). Geographically, Aleuritia is one of the most widespread groups within Primula, with main centers of diversity in the major mountain systems and the plains of North America and the major mountain systems of Eurasia, except for the Himalayas (Fig. 1a,c–e, Table 1). The only South American species of Primula described so far, Primula magellanica, belongs to Aleuritia (Fig. 1b, Table 1).
Table 1. Distributional range, latitude, ploidy level, breeding system and pollen type of Primula sect. Aleuritia species
Species included in the present study are in bold type. HE, heterostyle; HO, homostyle; na, not available.
Canada, south to Great Lakes region and west to central
2n = 2x = 18
P. farinosa L.
Alaska, England, Denmark, Sweden to 64° N, Finnish archipelago; Baltic states to 60° N, Montes Universales (Spain), Pyrenees, Alps, Tatra
2n = 2x = 18
P. anvilensis Kelso
Seward Peninsula, Alaska
2n = 2x = 18
P. incana M. E. Jones
2n = 6x = 54
P. halleri Gmel.
E Alps, Tatra, Carpathians, ex-Yugoslavia, Albania, Rila and Pirin ranges (Bulgaria)
2n = 4x = 36
P. yuparensis Takeda
2n = 4x = 36
P. magellanica Lehm.
Tierra del Fuego, Patagonia, Falkland Islands
2n = 8x = 72
P. laurentiana Fernald
2n = 8x = 72
P. borealis Duby
NW Alaska, N Canada, NE Siberia
2n = 4x = 36
P. scotica Hooker
2n = 6x = 54
P. scandinavica (Bruun) Bruun
Norway, a few sites in Sweden
2n = 8x= 72
P. stricta Hornem.
Scandinavia, eastward to Kola Peninsula (Russia), Novaya Zemlya (Russia), N Iceland, W Greenland, NE Canada
2n = 10–16x= c. 88–136
P. baldschuanica B. Fedstch.
Tadzhikistan, E Afghanistan
P. capitellata Boiss.
Iran, Afghanistan, Pakistan
P. sachalinensis Nakai
E Siberia, Kamtschatka and Sakhalin I.
P. schlagintweitiana Pax
N Pakistan, Kashmir and NW India
The Aleuritia polyploid system comprises at least five different ploidy levels, including the diploids Primula alcalina, Primula anvilensis, Primula farinosa, Primula frondosa, Primula mistassinica, Primula modesta and Primula specuicola (2n = 18), the tetraploids Primula borealis and Primula halleri (2n = 36), the hexaploids Primula incana and Primula scotica (2n = 54), the octoploids Primula laurentiana, P. magellanica and Primula scandinavica (2n = 72), and the 14-ploid Primula stricta (2n = c. 88–136; Table 1). Chromosome numbers are unavailable for four Asian endemics: Primula baldschuanica, Primula capitellata, Primula sachalinensis and Primula schlagintweitiana. All eight diploid species are heterostylous, whereas 10 of the 11 polyploid species are homostylous. The tetraploid P. borealis is heterostylous. Furthermore, the highest polyploids (8x−16x) rarely occur south of latitude 45° N or north of latitude 40° S, whereas the diploids usually occur between these latitudes (Fig. 1; Table 1). This geographic pattern suggested an association between high-latitudinal distribution, polyploidy, and homostyly (Bruun, 1932; Richards, 2002) and has been explained within the framework of the secondary contact model described above (Hultgård, 1990; Kelso, 1991, 1992; Hultgård, 1993).
Despite the publication of several recent papers on Primula (Hultgård, 1990; Kelso, 1991, 1992; Hultgård, 1993; Conti et al., 2000; Mast et al., 2001; Trift et al., 2002), no study has attempted to elucidate the evolutionary history and biogeography of Aleuritia within a detailed phylogenetic framework. Furthermore, while the proposed correlation between homostyly and polyploidy has been investigated in other genera such as Amsinckia, Damnacanthus and Turnera (Barrett & Shore, 1987; Schoen et al., 1997; Naiki & Nagamasu, 2004; Truyens et al., 2005), the combined effects of glacial cycles on these traits and biotic patterns of distribution has never been explored in light of an explicit phylogenetic hypothesis. Therefore, the present study intends to reconstruct the phylogenetic relationships within Aleuritia to (1) elucidate the biogeographic history of the European and American taxa; (2) infer whether ploidy levels and breeding systems changed once or rather multiple times, and (3) interpret the results in light of the selective forces that are likely to have modulated the interconnected evolution of ploidy levels, reproductive biology, and biotic ranges.
Materials and Methods
Taxon and locus sampling
A total of 63 plant accessions were used for the present study. Forty-four accessions represented 15 of the 21 species ascribed to Aleuritia and 19 accessions represented four of the 14 species ascribed to sect. Armerina (Richards, 2002). The Armerina accessions were included because cytological, morphological and biogeographic evidence suggested that Primula egaliksensis, assigned to sect. Armerina (Richards, 2002), might have a hybrid origin involving an Aleuritia and an Armerina parent (Kelso, 1991). Furthermore, a phylogeny of Primula based on two chloroplast introns supported the inclusion of selected members of Armerina and Aleuritia in the same, largely unresolved clade and, more specifically, of P. egaliksensis within a well-supported Aleuritia subclade (Mast et al., 2001). Finally, Primula ioessa (sect. Sikkimensis) was chosen to root the resulting tree because it was a member of the same large clade comprising Aleuritia and Armerina in the mentioned Primula phylogeny (Mast et al., 2001).
Sequences were generated from five noncoding regions of the chloroplast DNA (cpDNA), namely the rpl16 intron, rps16 intron, trnL intron, trnL-F spacer, and trnT-L spacer (see Supplementary material). The maternal inheritance of the cpDNA has been demonstrated in many angiosperms, including Primula (Corriveau & Coleman, 1988).
DNA extraction, PCR amplification, sequencing
Total genomic DNA was extracted using the DNeasy Plant Mini Kit (Qiagen, Hombrechtikon, Switzerland), after grinding the leaf material with glass beads in a Retsch MM 2000 shaker (Retsch, Haan, Germany).
Polymerase chain reactions (PCR) were performed in 20 µl volumes containing 1× buffer (including 1.5 mm MgCl2), 2 mm MgCl2, 200 µm dNTPs, 0.2 µm of each primer, and one unit Taq polymerase (Sigma, Buchs, Switzerland). Amplifications were carried out on a thermocycler (Biometra, Goettingen, Germany), using the following conditions: a first cycle at 94°C for 2 min; 35 cycles at 94°C for 30 s, 52°C for 1 min and 72°C for 1.75 min; and a final cycle of 10 min at 72°C. The rpl16 intron was amplified with primers rpL16F71 (5′-GCTATGCTTAGTGTGTGACTCGTTG-3′) and rpL16R1516 (5′-CCCTTCATTTCTTCCTCTATGTTG-3′; Small et al., 1998), the rps16 intron with primers rpS16F (5′-GTGGTAGAAAGCAACGTGCGACTT-3′) and rpS16R2 (5′-TCGGGATCGAACATCAATTGCAAC-3′; Oxelman et al., 1997), the trnL intron and trnL-F spacer (hereafter called trnL-F region) with primers trnL5′UAAF (TabC, 5′-CGAAATCGGTAGACGCTACG-3′) and trnFGAA (TabF, 5′-ATTTGAACTGGTGACACGAG-3′; Taberlet et al., 1991), and the trnT-L spacer with primers trnTUGUF (TabA, 5′-CATTACAAATGCGATGCTCT-3′) and 5′trnLUAAR (TabB, 5′-TCTACCGATTTCGCCATATC-3′; Taberlet et al., 1991). The length of the amplified fragments was estimated by comparison with DNA ladders on 1.2% agarose gels stained with ethidium bromide. The absence of contamination was checked by running out the PCR products of a negative control on the same gel. Successfully amplified PCR products (amplicons) were cleaned with the GFX PCR DNA and Gel Band Purification Kit (Biosciences Amersham, Otelfingen, Switzerland).
Sequencing reactions were prepared with the ABI Prism BigDye Terminator Cycle Sequencing Ready Reaction Kit (Applied Biosystems, Foster City, CA, USA), using the same primers as in the PCR amplifications. Sequencing products were purified on 96-well multiscreen filtration plates (Millipore, Billerica, MA, USA) to remove excess dye terminators, and run on an ABI Prism 3100 automated sequencer (Applied Biosystems, Foster City, CA, USA). sequencher 4.2 (Gene Codes Corp., Ann Arbor, MI, USA) was used to check the quality of the electropherograms and compile the contiguous sequences (contigs) for each amplicon.
The starting and ending points of each sequence were determined by comparison with the complete cpDNA sequence of Nicotiana tabacum (GenBank Z00044). Nucleotide sequences were aligned by eye (see Table 2, columns a). Unequivocally aligned gaps were coded as present or absent with the software gapcoder (Young & Healy, 2003), and added as binary characters to the end of the four-state nucleotide matrix (see Table 2, columns b).
Table 2. Summary statistics of the five chloroplast loci used for this study
Character metrics (a, matrix with nucleotides only; b, matrix with nucleotides and coded gaps): Total, total length of matrix (number of gaps); Informative, number of informative characters in the aligned matrix (number of informative gaps); Uninformative, number of uninformative characters in the aligned matrix (number of uninformative gaps); PICs, number of potentially informative characters (number of potentially informative gaps); % Variability, calculated by dividing the PICs value by the length of the aligned matrix. Tree statistics: number of trees (N); tree length (L); consistency index (CI); retention index (RI); and rescaled coefficiency index (RC). The effects of including gaps as additional characters on phylogenetic resolution are summarized at the end of the table.
Effects of including gaps:
Number of branches with higher BS
Each data set was first analyzed separately under maximum parsimony (MP) optimization. Visual comparisons of the trees derived from each data set independently revealed no strongly supported (bootstrap values > 59%) topological incongruence among the different cladograms. Therefore, data partitions were combined into a ‘global matrix’ that was analyzed with MP using the beta 10 version of paup* 4.0 (Swofford, 1999). Heuristic searches were performed with character states weighted equally, gaps treated as missing data and the following parameters: TBR branch swapping, Steepest Descent ON, Mulpars ON and Collapse branches option ON for branches with a minimum length of zero. Two-hundred searches were performed under these conditions, after randomizing the order of taxon addition. Five trees per replicate were saved and used as starting trees for a further round of branch swapping with TBR, now saving all the trees. Branch support evaluation was performed using the bootstrap method (Felsenstein, 1985). Bootstrap values (BS) were obtained from 100 000 fast-bootstrap replicates. This method is known to produce BS that are generally lower than those obtained using heuristic searches with stepwise addition and branch swapping (DeBry & Olmstead, 2000; Mort et al., 2000).
Conventional methods of phylogenetic reconstruction often produce low levels of resolution in cases of insufficient variability (Crandall, 1994), reticulation (e.g. hybridization), or multifurcation (i.e. single ancestral haplotypes giving rise to multiple descendants, Posada & Crandall, 2001). Under such circumstances, network analyses such as statistical parsimony, implemented in tcs 1.21 (Clement et al., 2000), can be profitably applied (Verheyen et al., 2003; Lihováet al., 2004). This analytical approach produces unrooted networks connecting only the haplotypes that have a high probability (> 0.95) of being similar due to shared history and not homoplasy (i.e. multiple hits; Templeton et al., 1992). Analyses were run on the global matrix, including all generated sequences, but excluding indels and missing data, which may create errors when collapsing sequences into haplotypes.
Patterns of character evolution were reconstructed using version 4.0 of macclade (Maddison & Maddison, 2000). To perform these analyses, the strict consensus tree of Fig. 2 was first pruned according to the following criteria: (1) when multiple accessions of the same species were monophyletic, only a single, randomly selected sequence was kept; (2) when multiple accessions of the same species were nonmonophyletic (see P. laurentiana and P. stricta), a single randomly selected sequence of each clade was kept; (3) P. halleri and P. scotica, which formed a polytomy and shared the same character states, were merged into a single taxonomic unit (halleri/scotica 1–2; see Fig. 4); (4) P. egaliksensis and two accessions of P. laurentiana, which formed a polytomy and shared the same character states, were merged into a single taxonomic unit (egaliksensis 1–8/laurentiana 3–4; see Fig. 4). Consequently, the tree used for character state reconstructions included 20 terminals. Ancestral state optimizations were performed on two topologies that differed in the position of P. magellanica, to reflect the different relationships of this taxon in the MP trees derived from the global matrix (data not shown).
The following two binary characters were evaluated: (1) breeding system (a. heterostyly; b. homostyly), and (2) ploidy level (a. diploidy; b. polyploidy). For character 1, the unordered states 1a and 1b were either equally weighted (i.e. a change from heterostyly to homostyly or vice versa had the same cost of one) or differentially weighted with a 2 : 1 ratio, favoring the loss of heterostyly over its gain. This weighting strategy was adopted because it has been argued that the origin of a complex trait, such as heterostyly, is likely to be less common than its loss (Kohn et al., 1996; Schoen et al., 1997). Similarly, we investigated the effects of using equal weights vs a 2 : 1 weighting ratio (favoring the shift from diploidy to polyploidy) on character state optimizations for character 2. The weighting ratio was selected because polyploids may become established (i.e. fully fertile) after only a few generations (Ramsey & Schemske, 2002), while supposedly requiring a much longer time span to get fully ‘diploidized’ again (Wendel, 2000; Leitch & Bennett, 2004). Both delayed (DELTRAN) and accelerated (ACCTRAN) transformations were used to resolve equivocal reconstructions.
To explicitly test the proposed correlation between polyploidy and homostyly within Aleuritia, we used the concentrated-changes test (CCT) implemented in macclade 4.0 (Maddison, 1990). This test evaluates whether changes in two binary characters are randomly dispersed on a tree or are concentrated along specific branches. More specifically, it tests whether gains or losses in a dependent character are more concentrated on the branches of an independent character than expected by chance (Maddison & Maddison, 2000). The probability of concentrated changes was assessed by exact counts in the Aleuritia clade for each reconstruction listed in Table 3. The same optimization scheme was used for both characters.
Table 3. Summary of character state reconstructions
Gains (a → b)
Losses (b → a)
Character 1 (breeding system): a, heterostyly; b, homostyly
1 : 1 ratio
1 : 1 ratio
2 : 1 ratio
No ambiguous reconstructions
Character 2 (ploidy level): a, diploidy; b, polyploidy
1 : 1 ratio
1 : 1 ratio
2 : 1 ratio
No ambiguous reconstructions
We designated character 2 (ploidy level) as the independent variable and character 1 (breeding system) as the dependent variable, based on the following rationale. In some plant taxa, polyploid genomes have been demonstrated to undergo rapid genomic rearrangements after their formation (Wendel, 2000; Soltis et al., 2003). This higher rate of recombination might increase the probability that the heterostyly linkage group is disrupted, thus explaining the higher frequency of homostyly in polyploids, as suggested by previous studies on Primula (Dowrick, 1956; Wedderburn & Richards, 1992). In this context, polyploidization would favor homostyly, justifying the selection of ploidy level as the independent variable. Alternatively, homostyly might have preceded polyploidization, but this hypothesis is contradicted by the absence of diploid homostyles and the occurrence of a polyploid heterostyle, P. borealis, in Aleuritia (Table 1).
The main characteristics of the five cpDNA regions used for the present study, along with the corresponding trees statistics, are summarized in Table 2. Parsimony analyses performed on the global data set, including gaps in the aligned matrix, produced four MP trees (length (L) = 335, consistency index (CI) = 0.928, retention index (RI) = 0.981, rescaled consistency index (RC) = 0.911; Fig. 2). The addition of coded gaps to the global data set resolved five additional clades and increased BS values for four branches, while causing a decrease only for the clade comprising all accessions of P. mistassinica (Table 2, Fig. 2). Some P. mistassinica accessions differ from each other by only one change, which is shared with one accession of P. laurentiana in a different clade (laurentiana 4).
The strict consensus of the four MP trees generated from the global data set supported the monophyly of sections Aleuritia and Armerina (BS 100%; Fig. 2), except for the placement of P. egaliksensis (taxonomically assigned to sect. Armerina) in the Aleuritia clade. Aleuritia can be subdivided into four main clades. Clade A (BS 92%), sister to the rest of Aleuritia, comprises four diploid species, P. modesta, P. specuicola, P. alcalina and P. anvilensis, and one tetraploid species, P. borealis. Clades B–D form a well-supported super-clade (BS 100%), but remain unresolved with respect to each other (Fig. 2). Clade B (BS 97%) comprises three accessions of the South American octoploid, P. magellanica (BS 97%). The poorly supported clade C (BS 63%) consists of several European and North American polyploids, namely P. stricta (14x), P. scandinavica (8x), P. incana (6x), and two accessions of P. laurentiana (8x; laurentiana 1–2). Finally, the main clade D (BS 98%) consists of a European and a North American subclade (subclades d1 and d2, respectively). The European subclade d1 is weakly supported (BS 59%), and the relationships among its taxa remain largely unresolved, as very low levels of cpDNA variation were detected between P. farinosa (2x), P. frondosa (2x), P. halleri (4x) and P. scotica (6x). Conversely, the North American subclade d2, formed by P. mistassinica (2x), P. egaliksensis (4x; sect. Armerina), and two accessions of P. laurentiana (8x; laurentiana 3–4), is strongly supported (BS 86%).
Seven species, for which multiple accessions were available, showed some degree of infraspecific variation. Two haplotypes were detected within P. scotica and P. egaliksensis, respectively, while three species, P. laurentiana, P. mistassinica and P. stricta, included three haplotypes each. Finally, five haplotypes were found within both P. farinosa and Primula nutans (sect. Armerina). While most infraspecific haplotypes differed only by a few nucleotides, the four haplotypes of P. laurentiana differed so much from each other that they were placed in unrelated clades (C and d2, respectively; Figs 2, 3).
The statistical parsimony analysis performed on the global matrix collapsed the sequences to 34 haplotypes and recovered two unlinked networks. The first network connected 21 haplotypes representing 45 accessions (Fig. 3). The haplotypes were linked into three major groups, corresponding to clades B, C and D of the MP consensus tree (see Fig. 2). Three mutational steps linked each one of these groups to the center of the network, which supposedly represents a missing ancestral haplotype connecting groups B, C and D to clade A (see Fig. 2). Group B connects the three identical accessions of P. magellanica; group C connects all accessions of P. incana, P. scandinavica and P. stricta along with two samples of P. laurentiana (laurentiana 1–2); group D connects all specimens of P. farinosa, P. frondosa, P. halleri, P. scotica, P. egaliksensis, and P. mistassinica, along with two identical accessions of P. laurentiana (laurentiana 3–4). Samples of P. farinosa collected in the central Alps (farinosa 1, 3) and England (farinosa 4–5), form the central haplotype of group D, called the ‘alpine’ haplotype from now on (Fig. 3).
The second network calculated by tcs (not shown) connected all P. nutans populations. The remaining haplotypes of the global matrix (i.e. all Armerina haplotypes, except P. nutans, and accessions belonging to clade A of Fig. 2) differed from the haplotypes of the first and second network by more than six steps and were not connected to either network because the null hypothesis that they were similar to each other due to chance (i.e. multiple hits) could not be rejected (P > 0.05).
The different position of P. magellanica in the two MP trees obtained from the reduced data set used for character state reconstructions had no effect on ancestral state optimizations for either character 1 (breeding system) or 2 (ploidy level). All reconstructions supported heterostyly as the ancestral state in Aleuritia, regardless of the weighting scheme or optimization strategy used. However, the number of times that homostyly was gained and lost changed with different weighting schemes and optimization strategies (Table 3). For example, under equal weighting (1 : 1 ratio), three gains and one loss of homostyly were reconstructed with ACCTRAN optimization, while four gains of homostyly were inferred when DELTRAN was used to resolve ambiguous optimizations. When shifts from heterostyly to homostyly were considered twice as probable as the alternative change, character state optimization for breeding systems was identical to that inferred under equal weights and DELTRAN (Fig. 4).
Similarly to heterostyly, all reconstructions supported diploidy as the ancestral state in Aleuritia, regardless of the weighting scheme or optimization strategy used. However, the number of switches from diploidy to polyploidy (i.e. gains) varied according to the applied weighting scheme and strategy used to resolve ambiguous optimizations (Table 3). Under equal weighting (1 : 1 ratio), four switches to polyploidy and one reversal to diploidy were reconstructed with ACCTRAN, whereas five polyploidization events were inferred with DELTRAN. Finally, when a 2 : 1 weight was employed, the pattern of character state reconstructions for ploidy level was identical to the one obtained with equal weights and DELTRAN (Fig. 4).
In the reconstructions shown in Fig. 4, both obtained with a 2 : 1 weighting scheme, four origins of homostyly were inferred along the five branches of the polyploid lineages, rejecting the null hypothesis that changes are distributed randomly (P < 0.05). Alternative reconstructions, which were inferred using an equal weighting scheme and ACCTRAN/DELTRAN optimizations (not shown), yielded similar results.
The phylogenetic results presented here provide novel evidence to explore the effects of Pleistocene glacial cycles on the evolution of biogeographic patterns, ploidy levels, and breeding systems in the Aleuritia diploid–polyploid species complex. The phylogeny, based on the maternally inherited chloroplast genome, also allows us to suggest the possible maternal lineages that contributed to the origins of polyploids in Aleuritia.
Biogeographic patterns and ploidy levels in Aleuritia
The cpDNA phylogeny confirms the monophyly of Aleuritia, including P. egaliksensis (taxonomically ascribed to sect. Armerina, Mast et al., 2001; Richards, 2002), and supports four main clades, discussed below (Figs 2 and 3). The sister group relationship of the mainly Asian Armerina to Aleuritia (Fig. 2), the topology of a genus-wide cpDNA phylogeny of Primula (Mast et al., 2001, 2006) and the observation that the Himalayan mountain system represents the main center of species diversity for Primula (Richards, 2002), suggest that an Asian ancestor possibly gave origin to an amphi-Pacific clade and to a lineage that diversified on the European and American continents. This lineage may have spread westward across the central Asian mountain ranges or via the arctic shorelines, giving origin to the amphi-Atlantic and European/North American clades, and south-eastward, giving origin to the sole South American species of Primula (P. magellanica). Clearly, a more rigorous interpretation of the geographic origin of Aleuritia will require an explicit reconstruction of ancestral areas of distribution at the genus-wide level. Furthermore, the haplotype network (Fig. 3) allowed us to identify the central biogeographic role of the ‘alpine’ haplotype of P. farinosa in Europe and suggests that it might have given origin to the North American accessions of clade D.
Amphi-Pacific lineage (clade A)
This strongly supported clade (BS 92%; Fig. 2), sister to the rest of Aleuritia, consists of four diploids (P. modesta, P. anvilensis, P. specuicola and P. alcalina) with very narrow ranges and one widespread tetraploid (P. borealis). The phenotypically variable P. modesta ranges from Japan to the Kurile Islands, with a few disjunct localities on the Korean mainland and neighboring Siberia (Fig. 1a). Primula anvilensis is restricted to alkaline soils of the Alaskan Seward Peninsula, P. specuicola to sandstone canyons of the south-western USA and P. alcalina to alkaline meadows of north-eastern Idaho (Fig. 1a). Such restricted and edaphically specialized distributions have been explained by affinity to cool, moist conditions that supposedly provided microrefugial habitats for these species (Kelso, 1987).
The contrast between the narrow distribution of diploids in clade A (Fig. 1a) and the broad distribution of most diploids in clade D (Fig. 1d–e) raises intriguing questions about the likely explanations of such different ranges for these heterostylous species. Part of the answer might be sought into the stricter edaphic requirements of P. alcalina, P. specuicola and P. anvilensis (Kelso, 1987). Additionally, the four diploid species of clade A currently occur in areas that were ice-free throughout the Pleistocene (Fig. 1a, Ono, 1985; Siegert, 2001). Consequently, these species may represent remnants of a more broadly distributed ancestral diploid that became fragmented during the advancement of Pleistocene glaciers and survived in refugial areas only (Kelso, 1991, 1992), as suggested for other North American plant groups (Soltis et al., 1997; Jaramillo-Correa et al., 2004; Godbout et al., 2005). Detailed comparative studies of the ecological characteristics of the diploid species in clades A and D, along with the evaluation of an absolute time-scale for their diversification, would be necessary to gain further insights into the likely origins of the observed distributional patterns.
The tetraploid P. borealis, unlike all other polyploids in Aleuritia, is heterostylous. This species displays a coastal, amphi-Beringian distribution from Japan to Alaska, expanding eastwards along the Canadian arctic shore, as far as the Mackenzie River delta (Fig. 1a). Morphological similarities, including rhomboid leaves and large flowers forming symmetrical umbels, prompted the suggestion of a close affinity between P. borealis and P. modesta, with non-overlapping ranges (Kelso, 1991, 1992). However, our maternal phylogeny supports a close relationship between P. borealis and the diploid P. anvilensis (BS 95%; Fig. 2), in agreement with present-day overlapping distributions (Fig. 1a). Currently, P. borealis occupies previously unglaciated areas, but its wider range supports the proposed higher colonizing success of polyploids compared with their diploid progenitors (Stebbins, 1985; Thompson & Lumaret, 1992).
South American lineage (clade B)
This clade comprises the three available accessions of the South American octoploid P. magellanica (BS 97%; Figs 2, 3), which ranges from Patagonia to Tierra del Fuego and the Falkland Islands (Fig. 1b). Morphologically, P. magellanica resembles two heavily farinose polyploids of North America, namely P. incana, with subcapitate inflorescences and flat bracts, and P. laurentiana, a robust species with large flowers. Our current data, based on cpDNA markers, do not support the conclusion that a P. incana- or P. laurentiana-like ancestor might have provided the maternal parent of this polyploid species, as previously suggested (Richards, 2002).
The phylogenetic position of P. magellanica remains unresolved (Fig. 2), indicating the need for additional data. The current topology of the strict consensus tree only allows us to suggest that it likely derived from the same common ancestor of clades C and D.
Amphi-Atlantic lineage (clade C)
This clade, comprising six North American and four European accessions of high ploidy levels (BS 63%, Fig. 2), is connected to the ancestral haplotype that gave origin to both the South American and the European/North American lineages (Fig. 3). Primula incana (6x) is restricted to inland clay soils of eastern Alaska, central and western Canada, and the Rocky Mountains, extending as far south as Utah and Colorado; P. laurentiana (8x) grows on shores and cliffs of north-eastern North America, expanding to western Newfoundland and the eastern shore of the Hudson Bay; P. scandinavica (8x) is endemic to Norway and part of Sweden; and the amphi-Atlantic P. stricta (14x) ranges from the eastern Canadian high Arctic, across Greenland, to northern Scandinavia and Russia, as far as Novaya Zemlya (Fig. 1c). The North American accessions of the amphi-Atlantic lineage are nested within the European ones (BS 86%; Fig. 2), suggesting that they were derived from European ancestors (see also Fig. 3).
High levels of cross-fertility between the European polyploid P. scandinavica and the diploid P. farinosa (Arnold & Richards, 1998) had prompted the suggestion that the former might be derived from differentiated populations of the latter (Bruun, 1932; Hultgård, 1990, 1993). A similar pattern had been proposed for the origins of P. incana and P. laurentiana from a diploid P. mistassinica-like ancestor on the North American continent (Kelso, 1991, 1992). While P. mistassinica and P. farinosa are both core diploid elements in North America and Europe, respectively (see the discussion on clade D later), the cpDNA phylogeny does not support a direct maternal descent of P. incana from a P. mistassinica-like ancestor or of P. scandinavica from a P. farinosa-like ancestor, because the hexaploid and octoploid taxa are included in a complex amphi-Atlantic clade composed entirely of polyploid elements (Figs 2 and 3).
As for P. laurentiana, our analyses point to independent origins of different populations ascribed to this taxon. Accessions of P. laurentiana collected on the Canadian mainland (laurentiana 3–4) are included in an unresolved polytomy with P. mistassinica and P. egaliksensis (subclade d2, BS 86%; Figs 2 and 3). Conversely, populations of P. laurentiana from Newfoundland (laurentiana 1–2) form a highly supported clade with P. incana (clade C, BS 95%; Figs 2, 3). The polyphyly of P. laurentiana s.l. and the additivity of its ploidy level (8x) with those of P. mistassinica (2x) and P. incana (6x) may imply that this octoploid derived from independent hybridization events involving either a P. mistassinica- or a P. incana-like ancestor as the maternal parent, respectively (Figs 2, 3). Recurrent origins have been demonstrated in other polyploid taxa (Soltis et al., 2003) and are congruent with proposed models for the evolutionary effects of repeated range fragmentation and expansion concomitant with the glacial cycles of the Pleistocene (Stebbins, 1984, 1985).
Based on chromosome number and cross-fertility evidence, two possible pairs of progenitors have been proposed for the 14-ploid P. stricta: the North American P. incana (6x) and P. laurentiana (8x), and the European P. scotica (6x) and P. scandinavica (8x, Kelso, 1991; Kelso, 1992; Arnold & Richards, 1998). The current overlapping distributions of P. stricta with P. laurentiana in North America and P. scandinavica in Europe further indicate a potential role of the last two species in the origin of the former (Fig. 1c). While all three sampled accessions of P. stricta were included in clade C, they did not form a monophyletic group (Fig. 2). The sole North American accession of P. stricta (stricta 3) formed a well-supported subclade with the North American P. incana and P. laurentiana (laurentiana 1–2, BS 86%), while the European accessions (stricta 1–2) were sister to the rest of clade C (BS 63%), which included one of the proposed European parents, P. scandinavica (Fig. 2). Therefore, different P. stricta populations might have independent origins from two different sets of progenitors on the two sides of the North Atlantic, providing yet another example of recurrent polyploid origins (Soltis et al., 2003). Analyses of nuclear markers might allow us to further clarify the origins of P. stricta.
European/North American lineage (clade D)
This clade (BS 98%; Figs 2 and 3), comprising both diploid heterostyles and polyploid homostyles, can be subdivided in two geographical groups. Sub-clade d1 (BS 59%) encompasses exclusively European species, whereas subclade d2 (BS 86%) encompasses North American species (Figs 2, 3).
The European subclade d1 (BS 59%; Figs 2, 3) may be regarded as a polyploid series centered on the diploid P. farinosa. Primula farinosa displays a disjunct distribution including the Spanish Sierra Nevada and Pyrenees, the Alps, the Carpathians, and the Baltic region (Fig. 1d). Primula frondosa (also 2x) is endemic to north-east Bulgaria, P. halleri (4x) to the eastern Alps and P. scotica (6x) to northern Scotland and nearby Orkney islands (Fig. 1d). Considering the widespread range and diploid state of P. farinosa, this species has been proposed as the ‘ancestor’ of most extant European taxa (Bruun, 1932; Hultgård, 1990, 1993). Different accessions of P. farinosa form a polytomy with the clade comprising P. frondosa, P. halleri and P. scotica (BS 63%; Fig. 2), thus not rejecting its proposed central role in Europe. Furthermore, the central position occupied by the ‘alpine’ haplotype of P. farinosa (farinosa 1, 3–5) in the corresponding part of the network (Fig. 3) suggests that it represents the ancestral haplotype of clade D (Crandall & Templeton, 1993) and that this lineage diversified from central Europe. The haplotype network further suggests that the North American lineage d2 might have derived from an ‘alpine’-like haplotype of P. farinosa.
The cpDNA sequences of P. farinosa are characterized by infraspecific polymorphism (Fig. 3). The five detected haplotypes are geographically structured, with two haplotypes in the Alps (farinosa 1, 3–5 and 7), two in the Pyrenees (farinosa 2 and 6), and one in the Carpathians (farinosa 8–9; Fig. 3), suggesting past isolation in separate refugia (Hewitt, 2004). Phylogeographic studies identified similar patterns of geographical structuring in the genetic diversity of Trollius europaeus (Després et al. 2002), Pritzelago alpina (Kropf et al. 2003) and Ranunculus glacialis (Schönswetter et al. 2003), supporting survival during glacial maxima in different southern refugia. Following glacial retreat, Northern Europe was likely recolonized by more southerly source populations either in the Alps or in the Carpathians. The haplotype shared by alpine (farinosa 1, 3) and British populations (farinosa 4–5) of P. farinosa (Fig. 3) denotes a possible migratory route between the central Alps and the British Isles. The investigation of Scandinavian populations of P. farinosa is likely to provide further evidence to elucidate postglacial migration routes between northern and southern Europe.
The diploid P. frondosa forms a weakly supported polytomic clade with P. halleri (4x) and P. scotica (6x, BS 63%; Fig. 2), suggesting that a P. frondosa-like ancestor might have been involved in the origin of either polyploid species. An allopolyploid origin for P. scotica had been suggested based on the high levels of allelic variation found in its populations (Glover & Abbott, 1995).
The diploid P. mistassinica in clade d2 (BS 86%; Figs 2 and 3) may constitute the North American counterpart of the European P. farinosa in clade d1 (Vogelmann, 1956; Kelso, 1991, 1992). This species is widely distributed across the boreal forest of North America (Fig. 1e), and forms a diploid–polyploid complex with P. egaliksensis (4x) and the Canadian accessions (laurentiana 3–4) of P. laurentiana (8x). Similarly to P. farinosa, the polymorphism detected in the cpDNA sequences of P. mistassinica, with one haplotype to the east (mistassinica 1–6) and two haplotypes to the west of the Great Lakes (mistassinica 7–8 and 9), might reflect the influence of Pleistocene glacial cycles on infraspecific genetic structuring (Fig. 3).
At least five glacial refugia have been identified in North America: (1) east of the Appalachians; (2) along the North Atlantic coast of Canada (Newfoundland); (3) in the Midwest, south of the Laurentide Ice Sheet; (4) in the west, south of the Cordilleran Ice Sheet; and (5) in the ice-free region of Beringia (Soltis et al., 1997; Siegert, 2001; Jaramillo-Correa et al., 2004; Godbout et al., 2005). More specifically, Vogelmann (1956) proposed the Appalachians, Newfoundland and the Midwest as potential refugia for P. mistassinica. Our results suggest that isolated populations of P. mistassinica might have survived in either the Appalachians or Newfoundland, giving origin to haplotype 1–6, while haplotypes 7–8 and 9 might be derived from relictual populations either in the Midwest, the west, or Beringia. Finally, the past presence of vast glacial lakes in central North America, especially Lake Agassiz between 13 000 and 8000 years ago, may account for the east–west partition of haplotypes in P. mistassinica (Siegert, 2001).
The tetraploid P. egaliksensis extends from the Russian Bering Sea coast to North America, Greenland, and Iceland, where it is likely to be extinct (H. Kristinsson, pers. comm.; Fig. 1e). This species has traditionally been placed in the primarily Asian section Armerina, based on similar morphological characteristics, including lack of farina, entire petiolate leaves, and narrow elongate capsules. However, the overlapping distributions of P. egaliksensis, P. mistassinica and to some extent P. nutans (sect. Armerina), a chromosome number for P. egaliksensis that is additive between diploids of sects Aleuritia and Armerina, and gland type, pollen size, colpi number and exine reticulations that are intermediate between sects Aleuritia and Armerina suggested that this tetraploid species might represent the product of hybridization between a P. mistassinica- and a P. nutans-like ancestor (Kelso, 1991, 1992). Recent phylogenetic analyses based on two chloroplast markers and single accessions indicated that the P. mistassinica lineage might have provided the maternal parent of P. egaliksensis (Mast et al., 2001). The present study, encompassing larger infra-specific sampling and three additional cpDNA markers, does not contradict the former phylogenetic conclusion, since all accessions of P. egaliksensis fall in the same clade with samples of P. mistassinica (BS 86%; Fig. 2).
Evolution of polyploid, homostylous species in Aleuritia
The increased rates of recombination that appear to be typical of some polyploids (Wendel, 2000; Soltis et al., 2003) might favor the disruption of the heterostyly linkage group (Dowrick, 1956; Wedderburn & Richards, 1992), thus providing a possible explanation for the frequent association between polyploidy and homostyly. The hypothesis that the switch from diploidy to polyploidy might favor the subsequent switch to homostyly is further supported by the fact that one Aleuritia-species is polyploid and heterostylous (P. borealis), whereas none is known to be diploid and homostylous (Table 1). At the same time, polyploidy may have reduced the potentially deleterious effects of inbreeding depression in self-compatible homostylous species, because polyploidization appears to be associated with an increase in genetic diversity (Lande & Schemske, 1985; Thompson & Lumaret, 1992).
It has been suggested that most Aleuritia polyploids are of hybrid origin (Bruun, 1932; Vogelmann, 1956; Hultgård, 1990; Kelso, 1991, 1992; Hultgård, 1993; Glover & Abbott, 1995). Studies on synthetic polyploids of Brassica showed that the frequency of genomic rearrangements in polyploids may be associated with the degree of genomic divergence between the diploid parents (Song et al., 1995; Wendel, 2000; Soltis et al., 2003). Therefore, allopolyploids, derived from interspecific hybridization, may experience higher rates of genomic rearrangements than autopolyploids derived from conspecific parents (Stebbins, 1950). Hence, the prevalence of homostyly in polyploid lineages of Aleuritia may be linked primarily to hybridization rather than chromosome doubling alone (see also Kelso, 1992). In this context, the maintenance of a heteromorphic breeding system in the tetraploid P. borealis raises interesting questions concerning the origin of polyploidy in this species, whether it occurred via allopolyploidy or autopolyploidy and whether it occurred recently or not.
In Aleuritia, polyploid, self-compatible homostyles are found at high latitudes (Table 1), so that a close association between distribution, ploidy levels, and breeding systems has been proposed (Bruun, 1932; Hultgård, 1990, 1993; Kelso, 1991, 1992; Richards, 2002). However, the diploid heterostylous P. anvilensis and the polyploid heterostylous P. borealis also occupy very high latitudes (Table 1), but in areas that probably remained ice-free throughout the Pleistocene (Fig. 1a, Ono, 1985; Siegert, 2001). In this regard, it is also useful to compare the distribution of heterostylous and homostylous species that now occur, partially or entirely, in previously glaciated areas on the same continent (Siegert, 2001), for example P. mistassinica (2x) with P. egaliksensis (4x), P. laurentiana (8x; Fig. 1e) and P. incana (6x; Fig. 1c) in North America, and P. farinosa (2x) with P. scotica (6x; Fig. 1d), P. scandinavica (8x) and P. stricta (14x; Fig. 1c) in Europe. This comparison suggests that the higher polyploid homostyles tend to expand further north into previously glaciated areas than the diploid heterostyles (see also Table 1), implying that the switch to autogamy may have represented a selective advantage in the colonization of barren postglacial habitats (Stebbins, 1957; Kelso, 1992).
Indeed, heterostylous populations of Aleuritia possibly growing at the ice-sheet margins may have suffered from mate deficiency, as predicted by the ‘abundant center model’, according to which fewer mates are available at the edge than at the center of a species’ range (Hengeveld & Haeck, 1982). The challenges caused by mate deficiency might have been compounded by the reduced pollinator activity (McCall & Primack, 1992; Totland, 1994) associated with very low temperatures during the glacial maxima (Siegert, 2001). Furthermore, insect visitation rates appear to be positively correlated with flower density; thus pollinator visits decrease at the edge of a species’ range (Thomson, 1981; Jacquemyn et al., 2002). Consequently, heterostylous, self-incompatible individuals, which depend on pollinators for successful fertilization, may have experienced lower reproductive success at the fringe of the Quaternary ice sheets resulting from mate and pollen limitation. Under such ecological conditions, selection for reproductive assurance may have favored the establishment of homostylous, self-compatible mutants that were independent of both mate density and pollinator activity (Stebbins, 1957; Fausto et al., 2001; Kalisz et al., 2004).
In general, the wider distribution of homostylous than heterostylous Aleuritia species in previously glaciated areas of high latitudes may reflect the higher colonization potential conferred by autogamy (Baker, 1955; Stebbins, 1957). Indeed, homostylous, self-compatible Aleuritia species were probably better adapted to invade habitats opened by glacier retreat, because only a few individuals would have been sufficient to found a new population (Baker, 1955; Stebbins, 1957). In this regard, it is relevant to note that the most disjunct species of Aleuritia (and indeed, of the entire Primula), P. magellanica, is a homostylous octoploid (Fig. 1b).
To summarize, the switch to homostyly played an important role in regulating the success of polyploid Aleuritia species, by allowing the newly arisen polyploids to self-fertilize and expand their range (Baker, 1955; Stebbins, 1957; Thompson & Lumaret, 1992). Therefore, the high frequency of polyploid, autogamous species in the Arctic may be the product of selection for increased selfing ability in habitats where pollination is unreliable (Molau, 1993).
For help with the collection of tissue samples we thank Torbjørn Alm, Alan Batten, Bruce Bennett, Sean Blaney, Christian Brochmann, Luc Brouillet, Edward Buyarski, Matthew Carlson, Mary Beth Cook, Reidar Elven, Pamela Eveleigh, Tupuna Kovanen, Wendy Maovlic, Michael Oldham, Carolyn Parker, Carl Roland, Peter Schönswetter, Heidi Solstad, Andreas Tribsch, all the colleagues listed in Supplementary material and the following herbaria: ACAD, ALA, ALAJ, ALTA, AMNH, CAFB, CAN, CDFN, DAO, GB, HAM, ICEL, LKHD, MT, NFLD, NHIC, NSPM, O, QFA, S, SASK, SCFQ, SLU, SMI, TRH, TROM, TRTE, S, UBC, UME, UNB, UQTR, UPS, UWPG, WAT, WIN, WLK. We would also like to acknowledge John Richards for general guidance on Primula, Péter Szövényi for help in the interpretation of network analyses and two anonymous reviewers for helpful comments on earlier versions of the manuscript. The Georges und Antoine Claraz-Schenkung and the Swiss Academy of Sciences funded A.G.'s field trips. This study was initially facilitated by grant no. 3100–061674.00/1 from the Swiss National Science Foundation and funds from the Office for Equal Opportunity of the University of Zurich to E.C.