Volume 93, Issue 3 p. 460-469
Systematics and Phytogeography
Free Access

Molecular phylogeography and hybridization in members of the circumpolar Potentilla sect. Niveae (Rosaceae)

Bente Eriksen

Corresponding Author

Bente Eriksen

Botanical Institute, Systematic Botany, Göteborg University, P.O. Box 461, 405 30 Göteborg, Sweden

Author for correspondence (e-mail: [email protected])Search for more papers by this author
Mats H Töpel

Mats H Töpel

Botanical Institute, Systematic Botany, Göteborg University, P.O. Box 461, 405 30 Göteborg, Sweden

Search for more papers by this author
First published: 01 March 2006
Citations: 11

We appreciate the logistic and financial support received from the Polar Research Secretariat on the Tundra Northwest expedition 1999. Thanks to K. Breen, University of British Columbia, Canada; K. Hedlund, University of Lund, Sweden; and M. Popp, University of Uppsala, Sweden for providing plant material. The study was sponsored by The Swedish Natural Science Research Council (DNr. B691/1999) and The Royal Society of Arts and Science in Göteborg. Thanks also to the reviewers for constructive criticism.


Glacial events and the formation of ice-free areas serving as refugia for plants and animals are important in shaping present patterns of genetic diversity in arctic areas. Beringia, situated in northeastern Russia and Alaska, has been pointed out as a major refugium. This study focuses on the historical biogeography of the circumpolar taxon Potentilla sect. Niveae. The taxonomy of the group is complex, most likely highly influenced by hybridization and apomixis. cpDNA microsatellites together with AFLP fragments were used to map the genetic variability in the section, from Beringia across the Canadian Arctic to Greenland. The data support the hypothesis that Beringia, as well as parts of adjacent arctic Canada, served as refugia during the Wisconsinan glaciation, and there is some evidence for a northern and a southern migration route out of Beringia. The hair type groups within sect. Niveae are more or less genetically distinct, and hybridization, especially with sect. Multifida, takes place. Haplotype diversity as well as frequency is at its maximum close to the Last Glacial Maximum ice cap edge. This pattern can be explained by merging of previously isolated refugia, by repeated extinction/colonization events close to the ice edge, and by hybridization among sympatric taxonomical lineages.

The diversity and distribution of the present-day arctic flora have been highly influenced by the climatic changes of the Quaternary. During periods of glaciation, certain areas remained ice-free and served as northern refugia for arctic and boreal biota, where taxa have long resided and evolved genomically (Abbott and Brochmann, 2003). In this paper we present a phylogeographic analysis based on species of Potentilla (Rosaceae) to test the hypothesis that Beringia served as a glacial refugium for these plants.

Beringia was first mentioned in 1937 by Swedish botanist Eric Hultén and was proposed to have served as a major arctic refugium during the Wisconsinan glaciation. Today it is regarded as being delimited by the river Lena (125° E) in northern Russia and the river Mackenzie (130° W) in northwestern North America (Abbott and Brochmann, 2003). Based on current plant distributions, Hultén also suggested that ice-free areas in the Canadian Arctic Archipelago and North Greenland were potential refugia during Quaternary glaciations (Hultén, 1937).

The Last Glacial Maximum (LGM) of the Wisconsinan glaciation is defined as occurring during a period of low global sea level, as well as during a time of relative climatic stability, ~18 14C ka BP (14C years are determined from isotope analysis; 18 14C ka corresponds to approximately 21000 years; Clark and Mix, 2002). Three major ice sheets were present in North America at LGM. The Laurentide Ice Sheet was the largest and had its center over Hudson Bay and adjacent central Canada. It started to build up at ~27–30 14C ka BP from an ice sheet that approximately followed the margin of the Canadian Shield (Fig. 1). In the northern part of the Canadian Arctic Archipelago, north of Parry Channel, the Innuitian Ice Sheet formed. The third ice sheet, the Cordilleran Ice Sheet, built up over the mountains of western Canada. These three independent ice sheets are considered as having more or less coalesced during the LGM (Clark and Mix, 2002; Dyke et al., 2002; Marshall et al., 2002; Fig. 1). Ice lobes were present in Amundsen Gulf to the south (Fig. 1, location 1) and in M'Clure Strait between Banks Island and Prince Patrick/Melville Islands farther north (Fig. 1, location 5), but Banks Island and Prince Patrick Island together with most of Melville Island remained unglaciated (Dyke et al., 2002). Hence, two distinct areas were separated from Beringia by ice streams, which may have acted as dispersal barriers between them and Beringia. The ice lobes lasted until approximately 13 14C ka BP (Dyke et al., 2002). Consequently, the western Canadian Islands formed additional refugia for the duration of the stadial. To the east, there may also have been nunataks along the coast of Ellesmere Island (England, 1999) and ice-free glacial foreland on Baffin Island (Miller et al., 2002).

As a result of the lowered sea level, the continental shelf between Asia and North America was exposed, allowing plants and animals to disperse across the continental border. The prevailing vegetation on the land bridge seems to have been mesic tundra (Elias et al., 1997; Alfimov and Berman, 2001).

Chloroplast DNA (cpDNA) markers have proved to be informative in phylogeographic studies (King and Ferris, 1998; Taberlet et al., 1998; Tremblay and Schoen, 1999; Abbott et al., 2000; Palmé and Vendramin, 2002). Because the chloroplast is usually maternally inherited, the alleles are dispersed only by seed. Seeds often have shorter dispersal distances than pollen, and the migration of alleles is therefore slower and easier to track. Hence, together with its low mutation rate, cpDNA is likely to retain historical structures such as past migration routes (Comes and Kadereit, 1998). In addition, effective population size for cpDNA is only half that for nuclear genes. Consequently, the probability for alleles to become fixed through random genetic drift is higher (Hartl and Clark, 1997). This may lead to increased differentiation among populations.

Thus, molecular genetics can describe intraspecific geographical structure by identifying lineages and consequently can reveal postglacial colonization routes provided that the location of the refugia are known (Taberlet et al., 1998). Based on the geological evidence of ice sheet dynamics, the location of the refugia can be predicted. Refugia typically harbor plants with a larger number of alleles than adjacent areas (Comes and Kadereit, 1998; Taberlet et al., 1998) because organisms within the refugia have had a longer time to accumulate mutations in the microsatellite region and migration is a random process in which only a subset of haplotypes will be dispersed to newly deglaciated areas. In different refugia, microsatellites will evolve independently, and the isolation leads to differences in haplotypes among regions. That means that even in present-day coherent populations of a species, the origin of single individuals can often be established. Molecular data on Saxifraga oppositifolia L. (Abbott et al., 2000) and Dryas integrifolia M. Vahl (Tremblay and Schoen, 1999) indicate that Beringia was indeed an important refugium during late Wisconsinan glaciation. Additional refugia may have existed in the northern parts of the Canadian Arctic Archipelago, just as there is some evidence that at least Dryas survived in the eastern coastal areas. Increased levels of diversity also occur in contact zones between migrants from different refugia (Abbott et al., 2000; Weider and Hobæk, 2003).

Amplified fragment length polymorphism (AFLP) is a technique based on differences in restriction sites across genomic DNA. In contrast to the cpDNA microsatellites, AFLP is able to show the sum of the parental genetic contributions to an individual. Individuals with a haplotype pertaining to one species, but a morphological expression indicating putative hybridity will be intermediate to the parents in a phenetic analysis based on AFLP data. In this study we have taken advantage of both techniques and their potential to answer questions about historical biogeography and hybridization.

The genus Potentilla L. (Rosaceae), comprising about 500 taxa worldwide (Mabberley, 1997), is a common element in arctic tundra ecosystems. The vast majority of species are perennial herbs. This study focuses on the circumpolar section Niveae consisting of species with trifoliate leaves, but for comparison we also treat P. pulchella R. Br. from sect. Multifidae with pinnate leaves.

Apomixis is known to occur in the genus Potentilla (Gustafsson, 1946, 1947; Asker and Jerling, 1992) but to what extent is a matter of debate (Eriksen, 1996; Holm and Ghatnekar, 1996a, b; Holm et al., 1997; Nyléhn and Hamre, 2002). Hybridization has been suggested as playing a major role in speciation within the genus (Soják, 1986), with equivocal evidence. Individuals deviating by having pentafoliate leaves, but otherwise conforming to sect. Niveae, have been interpreted as hybrids between taxa from sect. Niveae (trifoliate) and sect. Multifidae (pinnate). A large number of hybridigenous taxa were described by Soják (1986) and Yurtsev (1989) on these grounds. However, hybrid origin is not the only reason plants may grow pentafoliate leaves. Independent greenhouse trials on several taxa from sect. Niveae as well as the pentafoliate taxon P. insularis from sect. Multifidae showed that fertilization triggers production of supernumerary leaflets (Eriksen and Nyléhn, 1999). Obligatory apomixis is not compatible with the idea of hybridization, but if we accept that sexual reproduction does take place in many species (Holm and Ghatnekar, 1996a, b; Holm et al., 1997; Nyléhn and Hamre, 2002; B. Eriksen, personal observation), some hybridization might occur in areas where different taxa meet.

Probably as a consequence of hybridization and facultative apomixis, the taxonomy of sect. Niveae is complex. A total of 60 taxa have been described in the section, most of which are not very well defined. A morphometric study involving four Alaskan taxa of P. sect. Niveae showed that three main complexes can be identified based on their hair types as well as certain leaf characters (Eriksen, 1997). The accessions used in this study were classified into hair type groups according to these findings, and the discussions on historical biogeography in this paper relate to these species complexes rather than to named taxonomic entities.

The aims of the study

The objectives of this study are (1) to establish whether the Potentilla species complexes show a pattern of genetic variation supporting the hypothesis that Beringia could have been a refugium for Potentilla during the Wisconsinan glaciation and, thus, the source of present-day genotypes and (2) to map haplotypes from Beringia across the Canadian Arctic to Greenland, as well as to cluster AFLP genotypes in order to find evidence for postglacial migration routes.


Molecular analyses

Universal chloroplast primers developed by Weising and Gardner (1999) were used for this investigation. The primer sites flank regions of nucleotide repeats of variable length, so called microsatellites. The variations in length of each microsatellite locus are interpreted as alleles, and seen across a set of primers, the allele size combinations constitute haplotypes.

In a pilot study on a selection of the accessions used in this investigation, five of six tested universal primers were successfully amplified. Of these five, one turned out to amplify an invariable site. The remaining four primers (ccmp2, ccmp6, ccmp7, and ccmp10) were capable of organizing the material into groups small enough to provide a meaningful genetic signal and yet large enough to allow geographical interpretation.

Total genomic DNA from leaves dried with silica gel was extracted using a DNeasy plant mini kit (Qiagen, Valencia, California, USA) according to the enclosed protocol and adjusted to 10 ng/μL.

The cpDNA amplification reaction had a total volume of approximately 13 μL, consisting of 5.55 μL distilled water, 6.3 μL HotStarTaq Master Mix (Qiagen), 0.125 μL of each of the forward and reverse primers (20 μmol/L), and 1.0 μL DNA (10 ng/μL). The PCR amplifications were performed in a GeneAmp PCR System 9700 (Applied Biosystems, Foster City, California, USA). An initial 15-min denaturation at 95°C was followed by 25 cycles of 94°C for 30 s, annealing for 45 s at 50°C, and extension at 72°C for 1 min. The amplification ended with a final extension at 72°C for 10 min.

The fragment analysis was performed using a Beckman Coulter (Fullerton, California, USA) CEQ 8000 Genetic Analysis System. The PCR products were mixed (multiplexed), and the concentration was adjusted with distilled water to optimize the signal strength detected by the system. A volume of 1.5 μL PCR mix with 0.5 μL CEQ 400 sizer (Beckman Coulter) and 38.5 μL sample loading solution (SLS) were used for the analysis. An allele list was created by the fragment analysis module in CEQ 8000 Genetic Systems (version 5.0.360). The list was manually edited, and the highest peak for each dye was accepted as the true peak (i.e., the size of the allele).

A haplotype distribution map was constructed in ArcGIS 8.8 (Environmental Systems Research Institute [ESRI], Redlands, California, USA) for each hair type group. The size of the pies represents the number of accessions sampled at that locality, and the different patterns represent the different haplotypes. Haplotype frequencies were calculated as the relative number of haplotypes found in a hair type group at each location (see Table 1).

The AFLP procedure follows the manual from the AFLP Core Reagent kit (GIBCO BRL Life Technologies, Gaithersburg, Maryland, USA). Six primer pairs with different selective nucleotides (E-AAG/M-CAC, E-AAG/M-CAG, EAAG/M-CTG, E-AAG/M-CTT, E-ACC/M-CAT, E-ACC/M-CTG) were used. The last primer combination did not amplify well and was discarded. The PCR amplifications were performed in a GeneAmp PCR System 9700. Electrophoretic separation of the amplified fragments was carried out on ALFexpress instruments (Amersham Pharmacia Biotech, Uppsala, Sweden). All bands representing fragments with a length between 200 and 500–600 bp, depending on the primer pair, were scored. The number of bands entering the analysis ranged from 94 to 98 depending on the hair type group. Presence or absence of bands was registered manually and transferred to spread sheets as 1 and 0, respectively. The data were transformed into a distance matrix (Jaccard) by PAUP* version 4.0 (Swofford, 1998) and transferred to Le Progiciel R (Casgrain and Legendre, 4 accessed February 2004, free software program at http://www.bio.umontreal.ca/Casgrain/R/) for principal coordinates analyses. The accessions of Potentilla analyzed for AFLP are basically the same as were used for the haplotype study (see Appendix), and the haplotype of each sample is shown in the plots.

Plant material

A total of 203 accessions (Appendix) were used. Most material was identified as belonging to the circumpolar Potentilla sect. Niveae. This section is characterized by its trifoliate leaves covered by the white wool on the lower leaf surface and sometimes on the petiole. Three species complexes are recognized on account of their hair type (Eriksen, 1997). The P. uniflora complex has a mixture of crispate wool and long, smooth, straight hairs on the petioles. The P. nivea complex has a mixture of lanate wool and long, smooth, straight hairs, and the P. hookeriana complex has a mixture of lanate wool and long, verrucate straight hairs. For hair type morphology, see Eriksen and Yurtsev (1999). The five taxa belonging to the uniflora complex are P. uniflora, P. villosa, P. villosula Jurtz. (due to diffuse species limits this taxon is listed as P. uniflora), P. vahliana Lehm., and P. vulcanicola Juz. Adding to these is a number of collections matching the uniflora complex hair type but having pentafoliate leaves instead of trifoliate ones (Fig. 2b, c). The third complex, the hookeriana complex, comprises P. arenosa (Turcz.) Juz. (described from Siberia), P. hookeriana (described from USA), P. rubricaulis Lehm. (due to diffuse species limits this taxon is listed as P. hookeriana), and P. furcata Porsild. Among the latter collections are a number of pentafoliate individuals. The Appendix lists the accessions by the names commonly used in Potentilla taxonomy along with the hair type assigned to them.

In addition, a single species, P. pulchella, from section Multifidae was studied. The section Multifidae includes taxa with pinnate leaves, with or without wool. Potentilla pulchella is sympatric with the taxa belonging to section Niveae occurring over large geographic areas and may be a potential hybridization partner. A single collection is an offspring from an artificial hybridization between P. hyparctica from subgenus Dynamidium as the maternal species and P. hookeriana as the paternal species. Also sampled were P. hyparctica and collections belonging to P. anadyrensis Juz., one to P. stipularis L., and three to P. fragiformis Willd. These last three species all belong to distantly related sections and will not be discussed further.

The bulk of the material was collected during the Tundra Northwest 1999 expedition to the Canadian Arctic Archipelago arranged by the Swedish Polar Research Secretariat (Molau et al., 1999). Material from Alaska and Eastern Siberia was collected by B. Eriksen in the years 1994–1995. Additional specimens from Canada (Ellesmere Island, Yellowknife) and Greenland were received from colleagues. Material from Sweden was included in order to determine whether haplotypes are possibly circumpolar in their distribution. A list of localities and the number of samples taken at each of them is given in Table 1.


A morphometric analysis was carried out to verify that the characters used to identify taxonomic groups in a previous study on Alaskan material (Eriksen, 1997) are applicable to a larger data set and that haplotypes are generally restricted to particular taxonomic entities. Flowers and leaves from 156 accessions for which haplotypes were determined (marked by an asterisk in the Appendix) were available for measurements. In addition 16 accessions without haplotype data were included. A total of 10 characters were scored (for selection of characters, see Eriksen, 1997): petiole length, upper leaflet length, lower leaflet length, upper leaflet width, lower leaflet width, upper leaflet tooth number, lower leaflet tooth number, incision depth, petiolule length, and ovule number. A canonical discriminant analysis (CDA) was applied to the data using the software package SPSS 11.0.3 (SPSS for Mac OS X, 2004). Hair type was used as the discriminating factor.


Morphometric analysis

The CDA analysis confirms that hair types are indeed valid as indicators of gross morphology and corroborates the conclusions made earlier on a smaller Alaskan data set (Eriksen, 1997). The analysis resulted in three significant canonical variates (P < 0.001). The first two variates account for 94.9% of the variation among hair type groups. Wilk's lambda, which gives the proportion of the total variance in a CDA due to variation within groups (in this case hair types), is 0.041 (df = 30). The null hypothesis that all means across hair types are equal is therefore rejected (P < 0.001). The standardized coefficients for canonical variates are high for variables related to leaf tooth number and size along the first canonical axis (X, Fig. 3), and high for variables related to leaf size and shape along the second axis (Y). The total proportion of correctly reassigned individuals is 0.89. The confusion occurs mainly between the hookeriana and pulchella hair type groups as well as between the pulchella and uniflora groups. In Fig. 3 the pentafoliate individuals are highlighted to determine whether the putative hybrids also bear evidence of intermediacy in their morphological characters.

Chloroplast haplotype data

Thirty haplotypes were found and 15 of them were shared between two or more samples. Eight haplotypes were shared between 10 or more samples, together making up 82% of the material investigated. The highest level of allele richness was found in the microsatellite region amplified by primer pair ccmp6. Thirteen alleles were found in this region, compared to four or five in the other three regions. The alleles of ccmp6 fall into three distinct groups. One group of six alleles ranges between 112 bp and 121 bp, another group of three alleles ranges between 132 bp and 134 bp, and the third group of alleles ranges between 144 bp and 148 bp (Table 2). Thus, there is a gap of ca. 10 bp between the groups. The group with the longest fragments is dominated by accessions belonging to P. pulchella. The majority of these accessions are classified into haplotypes H2 and H3 (Fig. 4a). The shorter fragments are characteristic for section Niveae. The haplotypes H1, H5, H10, and H11 correspond to the uniflora hair type group (Fig. 4b). The majority of the accessions of that particular hair type (74%) are categorized as one of these haplotypes. H1 alone contains 47% of the accessions. An array of haplotypes representing only a few accessions (H10, H16, H24) is found in the eastern and central parts of the Canadian Archipelago but seem to be absent from Beringia. Haplotype H7 is characteristic for the hookeriana hair type group (Fig. 4c), in which it is present in 42% of the accessions. The hookeriana complex also shares haplotypes with P. pulchella and the nivea group. Specific for the two subspecies of P. nivea are mainly the haplotypes H4, H6, H13, and H15 (Fig. 4d). Haplotypes H6 and H13 are unique for the Swedish accessions.

The two dominating haplotypes in P. pulchella are H2 and H3. These haplotypes are also present in the uniflora and hookeriana hair type groups in areas where the taxa overlap geographically. The accessions in the uniflora and hookeriana groups with haplotype H2 or H3 are all more or less pentafoliate (Fig. 3).

Allelic richness: refugia and migration routes

The haplotype frequencies within three of the hair type groups (uniflora, pulchella, and hookeriana) were highest in the area around Banks Island (Table 1; Fig. 4). Potentilla nivea, which was not found in this area, shows a pattern of increasing frequency from Anadyr (western Beringia) to Ivvavik (northeastern Beringia).

The principal coordinates (PCO) analysis based on AFLP data for P. pulchella results in two rather distinct clusters (Fig. 5a). The first three principal coordinate axes explain ca. 17% of the variation. The haplotypes are mixed in the clusters, and when the individual identities are applied to the plot, there seems to be no geographical structure (Appendix), although accessions from a single locality tend to cluster together independent of haplotype.

The uniflora hair type group is split into three main clusters (Fig. 5b). Cluster I mainly represents accessions from Beringia, cluster II represents material from Beringia as well as adjacent parts of the Archipelago, and cluster III represents individuals with a southerly distribution across the Canadian Arctic. The first three principal coordinate axes explain ca. 10% of the total variation.

The hookeriana group likewise forms three AFLP clusters. The main haplotype, H7, is present in all (Fig. 5c). Many of the accessions, however, represent individuals that share their haplotype with P. pulchella and may be clustering together because they are hybrids. The artificial hybrid between P. hyparctica (H9) and P. hookeriana from Alaska (H7) cluster together with individuals from its parental P. hookeriana population in cluster II. The first three principal coordinate axes explain ca. 38% of the total variation.

The AFLP analysis for the nivea group also results in three clusters (Fig. 5d). Cluster I comprises mainly the Swedish material. Some of the odd haplotypes from the American Arctic group with them, but in general the Swedish plants seem to have been reproductively isolated from the American populations. The other two clusters consist of accessions found in both Russia and Alaska. Cluster III includes a couple of individuals with the most widespread haplotype (H4) but also contains accessions with a distinct haplotype from the Beringian part of the Canadian Mainland (Ivvavik) and some with haplotype H12, which is shared with hookeriana. The cluster does not seem to be naturally coherent, but more material is needed to be certain. Interestingly, some of the haplotypes that are shared with other hair type groups (H8 with uniflora and H9 with hyparctica/uniflora) fall outside the clusters. The first three principal coordinate axes explain ca. 21% of the total variation.


Hair type-specific haplotypes most numerous in Beringia

The data presented in this study clearly show a structured genetic signal in the material. The best signals for interpretation of migration routes are obtained from the uniflora hair type group, which is distributed across the area of interest and for which many samples are at hand. Most haplotypes in this group are found in areas unaffected by the latest glaciation (Figs. 1 and 4b), and the number of haplotypes present per site decreases with distance from these refugia. This is in accordance with the results presented for Saxifraga oppositifolia (Abbott et al., 2000), Dryas integrifolia (Tremblay and Schoen, 1999), and Daphnia pulex (Weider and Hobæk, 2003) and corroborates the hypothesis that Beringia served as a refugium for Potentilla during the Wisconsinan glaciation. When looking at individuals within a locality and analyzing the variety of AFLP clusters to which they pertain (Fig 5b), we see the same pattern as observed for the haplotypes: plants from the formerly glaciated Canadian islands have less genetic diversity than those from the refugial areas.

Haplotype H1 seems to have been especially successful in colonizing the eastern territories after the ice retreated. It is worth noting that there are two haplotypes, H10 and H16, that are only found in the formerly glaciated areas. There are several possible explanations for this current distribution: (1) by chance we did not sample haplotypes H10 and H16 in the Beringian and Banks areas; (2) the haplotypes evolved in postglacial times and later dispersed (H10); or (3) there was an additional refugium in the east, possibly on nunataks on Ellesmere Island (England, 1999) or along the coast of Baffin Island (Miller et al., 2002). Tremblay and Schoen (1999) pointed out that Dryas integrifolia probably inhabited a northern as well as an eastern refugium during the latest glaciation, a conclusion agreeing well with the third possibility.

AFLP data show that two separate dispersal routes out of Beringia may have been in use after the glaciation: one in the north and one in the south (Fig. 5b). One cluster is centered in Beringia and also includes individuals from northern Banks Island and Ellesmere Island. Another cluster comprises mainly individuals from southern localities. It is also interesting to see that the more restricted haplotypes H5 and H8, centered in Beringia, and H10, in the southern part of the Canadian Arctic, each project into one of the three main clusters (Fig. 5b). There seems to be a barrier between these populations that restricted mating. At the local scale, individuals are related although they carry different chloroplast haplotypes. Hence, sexual reproduction and genetic exchange appear to have occurred regularly.

Among the haplotypes pertaining to the hookeriana hair type group, H7 has a wide distribution (Fig. 4c). As in the previous group, most haplotypes occur in the refugial areas in Beringia and on Banks Island (Table 1). For the hookeriana hair type group data are lacking from the eastern sites, and it is not possible to make conclusions about migration.

Likewise, material from the nivea hair type group has been restricted to Beringia (except for the Swedish samples) (Fig. 4d). By far the largest number of haplotypes, although not the highest frequency (see Haplotype frequencies at maximum on the edge of the ice), is found in the Russian part of Beringia. The haplotype H4 has the largest distribution, occurring on the North Slope of Alaska as well as on the Chukotka Peninsula. South of the Yukon River, close to the Alaskan Range, two haplotypes different from the Russian ones are found. One of these is shared with P. uniflora, but the other is unique. In the maps of Hultén and Fries (1986), the Asian population extends into the North Slope of Alaska, whereas individuals further south apparently belong to a population ranging from Alaska along the Rocky Mountains southward to Colorado. Their hypothesis is supported by molecular evidence in the form of chloroplast haplotype, but morphologically the two groups are alike. The three accessions from central Alaska are situated on the edge of the uniflora cluster in the CDA plot, which is not surprising considering their haplotype, but they do not differ in any significant way from the northern material. It is also worth noting that the Swedish and Beringian accessions do not share haplotypes and form separate clusters in the ALFP analysis based on nuclear data, most likely due to long-time separation of the two stocks.

Potentilla pulchella has two widely distributed haplotypes (H2 and H3) in the area sampled, coexisting across large parts of the previously glaciated areas in the Canadian Arctic Archipelago (Fig. 4a). Although there is sympatry, H2 has a slightly more westerly and H3 a more easterly distribution. This may indicate that the two haplotypes originate from different refugia and have colonized the High Arctic from two different directions. However, the area from which P. pulchella is sampled for this investigation is not large enough to give firm evidence for this. More samples are needed from Beringia in the west and Ellesmere/Baffin islands and Greenland in the east. Potentilla pulchella differs ecologically from the other species sampled, being capable of surviving under very harsh conditions and often inhabiting High Arctic deserts almost devoid of vegetation (B. Eriksen, personal observation). This makes P. pulchella a possible inhabitant of the proposed glacial foreland of the LGM refugium on Baffin Island (Miller et al., 2002). In any case, AFLP data indicate that the two haplotypes interbreed when they are sympatric (Fig. 5a).

Haplotype frequencies at maximum on the edge of the ice

A consensus of data from all species tells us that the area with the most haplotypes coincides with the unglaciated areas and that Beringia and Banks Island most likely were glacial refugia for this group of species. One striking result is that the maximum haplotype frequencies, as well as allelic richness, are found on Banks Island, situated on the edge of the LGM ice margin, rather than at the center of the large Beringian refugium (Table 1). Low haplotype frequency is theoretically expected as a result of either (1) a bottleneck event, i.e., genetic depletion of the populations due to decreasing population sizes or (2) founder effect due to small population sizes and the accompanying genetic drift during colonization of new areas. The relatively low haplotype frequencies in Beringia could be explained by bottleneck events and the low frequencies in the eastern Canadian Arctic Archipelago by founder effects, but there may be additional explanations for the pattern of high haplotype frequency in Banks Island and adjacent areas. Banks Island and Prince Patrick/Melville Islands were allegedly separate entities for ca. 7000 years (Dyke et al., 2002) and must have served as additional refugia separated from, although very close to, Beringia. Due to the intermediate position of Banks Island, it may have constituted a postglacial contact zone for the different lineages that evolved in the three areas. Similar contact zones have been identified for European trees (Petit et al., 2003). Alternatively, as the ice margin advanced and retreated locally in accordance with changes in the climate, habitats were formed and wiped out several times, leading to repeated colonization/extinction events. This process may have had an effect on intraspecific evolution, making chance, or stochasticity, a more important parameter. New land can by chance be colonized by a genotype that would not be very successful in its original population, but due to less competition is able to survive and reproduce. Thus, some of the basic rules of evolution are set aside, and more genotypes are able to contribute to the total gene pool of the species.

Evidence of hybridization in the formation of pentafoliate plants

Another reason for the increased haplotype frequency in the area could be that it is especially rich in shared haplotypes interpreted as the result of hybridization events. Each hair type group has a number of unique haplotypes, indicating that the groups may have a common ancestry separated from other groups. It should be emphasized, however, that haplotype identity does not necessarily reflect ancestry. It may simply be the result of random processes. However, when morphological evidence correlates with hair types, and hair types in turn have consistent haplotypes, there is an increased possibility that the genetic signal carries phylogenetic information.

Among the data presented here, haplotypes otherwise characteristic for one group are sometimes found in another. This is the case in the uniflora and hookeriana hair type groups in which there are a number of accessions carrying one of the two main P. pulchella haplotypes (Fig. 4b and c). The interesting result from the combined analysis of haplotypes and morphological information is that the accessions having shared haplotypes are also pentafoliate. In Fig. 3 the pentafoliate individuals are highlighted to determine whether the putative hybrids also bear evidence of intermediacy in their morphological characters. In the case of pentafoliate individuals with uniflora hair type, there is good morphological evidence of hybridization. The majority of the pentafoliate individuals fall in the region between the uniflora and the pulchella clusters. The coincidence of molecular and morphological data strongly suggests that these accessions are actually hybrids between the two hair type groups.

For pentafoliate members of the hookeriana group, the picture is not as clear. There are four data points of which three represent individuals with haplotype H3, which is the one shared with P. pulchella. The fourth point represents an individual with the normal hookeriana haplotype H7. The latter fall in the center of the hookeriana cluster and may be pentafoliate because of ecological factors. Among the other three, one point falls in the region between the hookeriana and pulchella clusters and is interpreted as a putative hybrid between the two taxa. The remaining two individuals are morphologically close to hookeriana and originate from Yellowknife, a region where P. pulchella does not currently and probably never did occur. Yet, the two individuals carry haplotype H3. We are not able to explain that directly, but one possibility is that the microsatellites by chance evolved allele sizes identical to those of pulchella. Another possibility is that the individuals are hybrids and have approached the morphology of one of their parents, P. hookeriana, by introgression.

Likewise, a few individuals in the nivea hair type group share haplotypes with individuals in the uniflora group (Fig. 4b and d). In this case, there are no simple morphological means of telling whether this is due to hybridization.

As a small test of the methodology, an artificial hybrid between P. hyparctica and P. hookeriana was included in the analysis. A number of pure P. hyparctica individuals were also investigated, although they are not presented otherwise in this work. The artificial hybrid is an offspring from a P. hyparctica maternal plant, emasculated and cross-pollinated by P. hookeriana pollen. The paternal plant originates from the population at Munson Slough (MU) in Alaska. Both species have trifoliate leaves, but P. hyparctica does not have wool on its leaves as does P. hookeriana. The offspring has a wooly indument showing that hybridization took place. As expected, the hybrid inherited its mother's haplotype (H9, Appendix). In the AFLP plot of the hookeriana group (Fig. 5c), the hybrid inserts neatly together with the paternal population (H7) in the central cluster.

Historical biogeography and evolution of Potentilla in Beringia

Beringia and associated ice-free areas on the western Canadian islands most likely served as refugia for Potentilla during the Wisconsinan glaciation. It is also plausible that glacial foreland or nunataks further east harbored species of Potentilla during the last glacial maximum. A postglacial dispersal from west to east seems to have taken place, possibly following a southern and northern route. It is evident that the hair type groups within the section Niveae are more or less genetically distinct and that hybridization, especially with P. pulchella, a member of section Multifida, takes place. Haplotype diversity as well as haplotype frequencies are at their maximum in the area close to the LGM ice sheet edge. This pattern can be explained by the presence of several isolated refugia in the western Canadian islands, by repeated extinction/colonization events close to the edge of the ice, and by hybridization among sympatric taxonomical lineages.

Table 1. The number of Potentilla samples taken at each locality for each hair type group. uni, a mixture of crispate wool and long, smooth, straight hairs on the petioles; niv, a mixture of lanate wool and long, smooth, straight hairs; hook, a mixture of lanate wool and long, verrucate straight hairs; pulc, hair type as uni but the leaf pinnate; hyp, lacking wooly indument. The number of haplotypes and the haplotype frequencies, calculated as the number of haplotypes per sample, are also given
Table 2. A list of the combinations of alleles that make up the haplotypes H1–H30 used to classify specimens of arctic Potentilla. C2–C10 refer to the primers (ccmp2–ccmp10) used to amplify the different fragments. Numbers represent the length of the fragments in base pairs
Details are in the caption following the image

Map of arctic region with probable ice distribution at 27–30 14C ka BP (white areas) and probable ice margin at the Last Glacial Maximum (LGM; dark line). The Beringian area, large parts of Banks Island, as well as parts of Melville and Prince Patrick Islands are considered to have remained ice-free during the entire Wisconsinan glaciation. 1, Amundsen Gulf; 2, Baffin Island; 3, Banks Island; 4, Ellesmere Island; 5, M'Clure Strait; 6, Melville Island; 7, Parry Channel; 8, Prince Patrick Island; 9, Victoria Island. Figure redrawn after Dyke et al. (2002)

Details are in the caption following the image

Leaf morphology for three sympatric Potentilla sections. (a) P. pulchella, (c) P. uniflora, and (b) their putative pentafoliate hybrid. The leaves were sampled at the same locality on Victoria Island. The pentafoliate individual has haplotype H2, as does P. pulchella. The specimen of P. uniflora has haplotype H11. It is evident that classification based on leaf morphology is rather difficult. Vouchers: (a) BE940:1CA, (b) BE924:3CA, and (c) BE980:1CA

Details are in the caption following the image

Three-dimensional scatterplot of canonical scores for the first three axes (canonical discriminant analysis, CDA) based on morphometric data for specimens of arctic Potentilla. The grouping factor of the CDA is hair type. Pentafoliate individuals, which are putative hybrids, are highlighted by a cross

Details are in the caption following the image

Haplotype distribution maps for each hair type group within Potentilla sect. Niveae and P. sect. Multifida. (a) pulchella, (b) uniflora, (c) hookeriana, and (d) nivea. The size of the marker indicates the number of samples taken at each locality. The patterns represent different haplotypes and are explained in the legend

Details are in the caption following the image

Three-dimensional scatterplot of canonical scores for the first three axes (principal coordinates, PCO) based on amplified fragment length polymorphism (AFLP) data for specimens of arctic Potentilla. Clusters (I, II, III) projecting in different directions are encircled. The symbols, representing haplotypes, are explained in the inserted legends