Phylogeography of a tough rock survivor in European dry grasslands

Phylogeographic analyses of plants in Europe have revealed common glacial refugia and migration routes for several trees and herbs with arctic-alpine distributions. The postglacial histories of dry grassland species in central Europe have rarely been analyzed, even though the extremely species-rich habitat is threatened. Sedum album (Crassulaceae) is a common inhabitant of rocky sites in central European dry grasslands. We inferred the phylogeographic history of S. album over its distribution range in Europe. Genetic diversity within and differentiation between 34 S. album populations was examined using AFLP markers. Population isolation was indicated based on the rarity of the fragments and by isolation-by-distance effects. We sequenced the trnL-trnF region in 32 populations and used chloroplast microsatellites to analyze chloroplast haplotype distributions. Two distinct S. album lineages were detected. One lineage was comprised of populations from eastern and central parts of central Europe, and the Apennine Peninsula. A second lineage was comprised of populations from the Iberian Peninsula and western and northern parts of central Europe. Glacial refugia were identified based on the accumulation of ancient chloroplast haplotypes, high diversity of AFLP fragments within populations, and high levels of rare fragments in Liguria, Serbia, the Apennine and Iberian peninsulas. Cryptic refugia were detected in the Czech Republic and Slovakia. Isolation by distance was present all over the distribution range, and it was separately detected in southwestern and central Europe. In western Europe, where a contact zone between the two lineages can be expected, no isolation by distance was detected. Our results suggest migration routes of S. album northeastward from glacial refugia in southern Iberia, northward from the Apennine Peninsula, and northward and westward from the southeastern parts of central Europe. Therefore, central European grasslands were recently colonized by northern cryptic populations and source populations originating in the east and the Apennine Peninsula.


Introduction
The geographical distribution of plant species has always been influenced by climatic and edaphic factors. Quaternary ice ages dramatically reduced the habitats of most species [1][2][3]. Large parts of northern Europe, the Alps, and the Pyrenees were covered by ice, and steppe- tundra habitats with cold and dry conditions dominated the vegetation in most parts of western and central Europe. Most species were only able to survive in southern and eastern Europe where climatic conditions were milder than those in northern parts. Nevertheless, next to the general contradiction-expansion theories, the plants and animals could have survived in northern areas characterized by suitable microclimate [4]. For many species, cryptic refugia in areas with sheltered topographies (e.g. river valleys and caves in river valleys) have been described in studies that are more recent [5][6][7].
Comparative phylogeographic studies identified patterns in the distribution of genetic diversity that could explain collective areas of glacial refugia and migration routes in Eurasian areas [3,8]. The Iberian Peninsula harbors the glacial refugia of several Eurasian species, and trees and shrubs from this region headed northward and westward after the last glacial maxima (LGM) [9][10][11][12]. Additional potential glacial refugia were identified for the genus Helianthemum in southern France and near the Pyrenees, from where the species migrated northward and westward [13]. Refugia were also suggested in Italy and on the Balkan Peninsula for several species [9,12,[14][15][16], which expanded northward and westward to central Europe. Mountain ranges, particularly the Alps and the Pyrenees, form natural barriers to the migration routes of European plant species [11,17,18]. Rivers can form barriers and-more likely for most plantsserve as transportation routes for dispersal or as cryptic refugia [5,19].
In addition to archeobotanical findings (most notably pollen depositions but also macrorests) genetic techniques are the methods of choice used to reveal the postglacial history of plant species [20]. Possible colonization scenarios are based on high haplotype diversity in glacial refugia. In areas that were recolonized after the LGM, diversity losses resulting from founder effects during rapid expansion have been suggested [1,21]. High genetic diversity within and strong differentiation between populations in southern areas (compared to that in the northern areas) are suggestive of the northward expansion of species [9,17,18,[22][23][24][25][26]. Nonetheless, high genetic diversity can also be characteristic for contact zones where different lineages from distinct genetic sources mix [27,28]. Therefore, landscape history has to be taken into account to interpret the genetic pattern of plant populations successfully. Stable populations with long histories on-site (e.g., in glacial refugia) further indicate an accumulation of rare haplotypes, which are good indicators of refugial areas [29]. Because of stochastic processes, the identification of rare haplotypes is less likely in recently colonized habitats [30].
Regarding migrations during interglacial periods and after the LGM, mankind was an influential factor for most plant species in central Europe [31,32]. Natural postglacial reforestation was restrained by land use (e.g., grazing and agriculture), and suitable habitats for thermophilic plants were formed, including dry grasslands that evolved on sites with grazing livestock. Several plants further dispersed via migrations and human settlement after the LGM, during which humans took plants or seeds with them-intentionally or unintentionally with pastoralism [33][34][35]. In recent studies, the postglacial migration routes of anthropochorous species (plants and animals) have been linked to human migrations (e.g., migrations with livestock) [31,[36][37][38].
The succulent Sedum album L. (Crassulaceae) is a typical inhabitant of dry grasslands in alkaline and calcareous soils in central Europe. It can be found on rocks and shallow soils within the Festuco-Brometalia vegetation type and in pioneer vegetation on rocks within the Sedo-Scleranthetalia-group (Sedo albi-Veronicion dillenii) [54][55][56]. S. album can also be found in crevices on stony riverbanks and in forests with oak or olea in the Mediterranean. Further, the species immigrates to man-made habitats, including the stone walls of buildings and gardens in cities and villages [33, [57][58][59]. Nowadays, S. album is common, and it is distributed over wide regions in temperate Europe and the Mediterranean.
In the present study, we investigated 34 populations of S. album in Europe using AFLP markers, chloroplast sequences (trnL-trnF), and chloroplast microsatellites to answer the following questions: 1) Can we infer the postglacial history of a common plant like S. album using standard methods (AFLPs and chloroplast markers)? 2) Was S. album introduced to central Europe after the LGM from glacial refugia in the southwest or southeast? 3) Can we detect cryptic refugia in central Europe? 4) What are the characteristics of the migration routes of this species to dry grasslands in central Europe?

Species and collection
Sedum album L. (Crassulaceae) is a perennial herb, with succulent woody stem and succulent, linear-cylindrical to ovoid leaves. Seeds are long-term persistent and dispersed by rain and water. Shoots and leaves are very resistant to drought and can develop adventitious roots when they encounter favorable conditions or habitats. The flowers have white to pink petals and are either pollinated by insects, or self-pollinated.
S. album was used as ornamental, medicinal, and edible plant, and was proven to grow near human settlements [33, [59][60][61][62][63][64]. The natural distribution range is all over Europe except for northern and eastern parts [55,65]. The eastern border of its distribution in Poland, Slovakia, Hungary, Romania, Bulgaria, and Turkey have been described [54,55]. In the south, this species can be found until northern Morocco, Algeria, and Tunisia. In the north the natural distribution is described in southern Sweden and Norway, whereas it was introduced in most parts of the UK and Ireland, and in parts of Sweden [55].
For the present study, 34 populations of S. album were sampled in dry grassland (Festuco-Brometalia) and rock slopes (Sedo albii-Veronicion dilenii) all over Europe (Table 1, S1 Table) except for the northern and eastern parts, where the populations might have been introduced [65]. S. album is not protected and we did not collect samples from protected areas, therefore no specific permissions were required. We avoided populations with characteristics of segregates, as well as populations near gardens and roof greenings to exclude the direct influence of horticultural S. album breeds. Fresh and young branches of six (GB) to 14 (S12) plants per population were sampled (Table 1) and stored in filter bags with silica gel until the material was frozen at -18˚C. Voucher specimens are deposited at the Institute of Plant Sciences, University of Regensburg.

DNA extraction
Approximately 5-8 leaves of each plant were homogenized with liquid nitrogen and DNA was extracted using the CTAB method [69] with minor adaptations [44]. DNA concentrations of all extractions were measured as transmission of optical density. A standard DNA concentration (7.8 ng/μl) was diluted, which was taken for all further investigations. AFLP analysis was conducted with 388 individuals in total following the protocol of Vos et al. [70] with minor variations [46]. About 50 ng of DNA was digested with EcoRI and MseI restriction enzymes (Fermentas) and ligated to adaptors (MWG) with T4 DNA Ligase Table 1. 34 investigated populations of S. album in Europe. Given are identification (ID), country, number of individuals for AFLP (nr), microsatellite haplotypes (ccmp), trnL-trnF sequence haplotype (seq), groups revealed by AFLPs (AFLP): Central Europe East (CE-E) and West (CE-W), Western Europe (WE), Iberian Peninsula (IP), Southern Italy and western Alps (S-It & Al), Liguria (Lig); further, percentage of polymorphic fragments (%PL), diversity as Shannon Index (SI) and SSWP/n-1 (SSWP), down-weighted rare fragments (DW), private fragments (priv) and fragments restricted to two populations (with) are shown. (Fermentas) at 37˚C for 2 h with a final step of 70˚C for 15 min. Restriction-ligation and all following polymerase chain reactions were carried out in automated Thermocycler (Eppendorf) with Taq Polymerase (PeqLab). The first amplification step was conducted with a pair of onebase primers (MseI-C/EcoRI-A, MWG) in 30 cycles. Fluorescence labeled Mse primers were used for the final amplification with three-base primers (MseI-C/EcoRI-A, MseI-C/EcoRI-A, MseI-C/EcoRI-A, MWG); 25 cycles were run of which 10 cycles had a 1˚C touchdown profile. The PCR products were precipitated with NaAc, EDTA, glycogen, and cold EtOH (96%, -20˚C) and cleaned with cold EtOH (70%, -20˚C). After the DNA pellet was dried, it was dissolved in Sample Loading Solution (Beckman Coulter) and charged with CEQ Size Standard 400 (Beckman Coulter). The capillary gel electrophoresis was run on an automated sequencer (CEQ 8000, Beckman Coulter) and the raw data were automatically analyzed with CEQ 8000 software (Beckman Coulter) for aligning the fragments according to CEQ Size Standard 400. Clear and well-defined fragments were scored manually in the program BioNumerics v. 3.0 (Applied Maths). The fragments with sizes of 60-420 bp were processed and scored across all individuals as either present or absent. Both individuals and fragment sizes that did not give clear, defined, and reproducible fragments were excluded from the analysis. The resulting binary matrix was used for further statistical analysis.

ID
To analyze the quality and reproducibility of the fragments, 55 individuals were replicated, which gave an error rate of 1.36% [71].
Genetic diversity within populations was calculated based on all markers that produced polymorphic fragments, percentage of polymorphic fragments, and Shannon Index (SI = ∑ pi ln pi) for each population [72], using the program POPGENE version 1.32 [73]. Sum of squares within populations (SSWP) was derived by analysis of molecular variance (AMOVA) in GenAlEx version 5 [74]. SSWP was divided by n-1 for each population to describe within population diversity independent from the number of individuals [75].
Rarity of fragments can indicate the level of isolation and differentiation [76]. Therefore, we tested populations for private fragments or shared fragments between two populations. We detected rarity as frequency-down-weighted markers (DW) for each population [30] with AFLPdat in R [77]. We randomly chose nine individuals per population in five iterations. Owing to few individuals, the GB population was excluded from this analysis.
Among population distances were estimated with the program AFLP-SURV [78] based on Nei's standard (Ds) distance with non-uniform prior distribution of allele frequencies. Based on Ds distances, we constructed consensus neighbor-net graphs with Splits Tree [79].
The differentiation within populations, among populations, and among main groups revealed by neighbor-net graphs and Bayesian clustering was calculated in GenAlEx version 5 [74]. Based on pairwise genetic distances between individuals, population differentiation was inferred as PhiPT, which is an analogue of Wright's F-statistics [80]. Further a hierarchical AMOVA was used to estimate differentiation between populations, within groups, and between groups [81].
Based on pairwise PhiPT values from AMOVA and pairwise geographic population distances (km) we conducted a Mantel-Test [82] in GenAlEx version 5 [74]. Correlations of genetic and geographic distance were tested using 999 permutations.
Genetically homogenous groups were inferred with Bayesian clustering in the program STRUCTURE version 2.2 for all individuals (K = 1-35) and two subgroups in separate runs (east: K = 1-21; west: K = 1-15) [83]. Allele frequencies were assumed to be correlated in an admixture model [84]. The number of groups was estimated using 10 5 iterations, with a burnin-period of 10 4 . For each predefined number of K, 10 iterations were run. From the resulting values of L(K) standard deviations and ΔK were calculated to affirm the most likely number of groups with STRUCTURE harvester [84,85]. The software CLUMPP was used to calculate similarity coefficients (H') between runs for the same K [86], and the output was used to visualize assignments of individuals to groups using the software DISTRUCT [87].

cpDNA sequence analysis
The plastid trnL intron and trnL-F spacer sequences were chosen for a preliminary test using primers Tab c and Tab f [88], because they were the most variable regions of the chloroplast genome. TrnS-G intron [89] and rpl32-trnL(UAG) [90] were tested, but excluded for further analysis owing to low variability or difficult amplification.
Two individuals per population were used for this analysis. All the samples were amplified in both directions (with Tab f and Tab c), but an overlap could not be achieved in any of the samples. To reassure the stability of the sequences, individuals were sequenced twice and consensus sequences were used for further analyses. Populations F5 from France and GB from Great Britain were excluded from further analyses because we failed to produce two appropriate sequences per population.
We compared the resulting sequences with those in GenBank using Blast [91]. Alignment was conducted using Clustal Omega [92] and adjusted manually with BioEdit Sequence Alignment Editor 7.2.0 [93]. The sequences were submitted to Genbank (S2 Table).
Concatenated cpDNA sequences were used to build a network under statistical parsimony (as implemented in Clement et al. [94]) in TCS v1.21 with gaps coded as missing. The connection limit was set at 120 steps to include the out-group Sedum dasyphyllum from Barcelona (BCN) [95].
Maximum parsimony (MP) heuristic searches were conducted in PAUP Ã 4.0a [96] with TBR branch swapping in action and gaps coded as missing. Confidence was assessed by bootstrap analysis with 1000 bootstrap replicates and 100 random addition sequence replicates per bootstrap (with a limit of 10 sec per sequence replicate).
For each individual, 1 μl of the PCR products of each primer was mixed and added to Sample Loading Solution, charged with an internal standard (CEQ Size Standard 400) and run on an automated capillary sequencer (CEQ 8000, Beckman Coulter). Chloroplast haplotypes were defined by fragment lengths using CEQ 8000 software. A haplotype network was built under statistical parsimony in TCS v1.21 [94]. Geographic sampling regions were used to analyze group structure of sequence and microsatellite haplotypes.  (Table 1). Genetic diversity ranged from SI = 0.1724 (E3) to SI = 0.391 (D1). Populations from the southern regions showed higher diversity than those from the northern and central Europe (Fig 1A). Likewise, the rarity of fragments was high in the southern populations (max. DW = 17.19 in I6 and max. private fragments = 6 in PO) and lower in the northern and the central populations (min. DW = 6.87 in I2) (Fig 1B). Private fragments were accumulated in populations D1 (Germany), I5 and I6 (Liguria), PO (Portugal), E4 (northern Spain), and I8 (Central Italy).

AFLP analysis
In the neighbor-net graph (Fig 2) the separation of southwestern Europe (Spain and Portugal), western Europe, Serbia, Liguria, and central Europe with Italy was evident. Central Europe and Italy formed one group and a further division in three subgroups was possible: central Europe-west consisted of populations from western Germany and Switzerland, Central Europe-east consisted of populations from Austria, Czech Republic, Germany, Italy, Slovakia, but also Great Britain. With the exception of the population from Great Britain, populations formed one lineage from northeast Italy northward to Germany and eastward to Czech Republic and Slovakia. Populations from the middle of the Apennine Peninsula and the western Alps (Italy and Switzerland) formed a third group. Populations I5 and I6 from northwestern Italy (Liguria) were segregated from the rest of the Italian group and clustered between the populations from Serbia and those from Iberia. D3 from Central Europe was separated in the network from its geographic neighbors and situated between the populations from Serbia and western central Europe.
Differentiation was high between populations in general (72.3%, Table 2). We found 32.7% differentiation between groups, which have been determined by neighbor-net analysis and were mainly congruent with geographic regions. We omitted the regions Liguria and Serbia (incl. D3) from the AMOVA and population structure analysis, because those regions contain a large number of STRUCTURE groups (Fig 3) and the assignment to only one group is, therefore, not accurate. The AMOVA of STRUCTURE groups revealed similarly high differentiation between both groups (29.8%).
To reveal the influence of other regions on central European populations, we tested differentiations of all combinations of the regions against each other ( Table 2). The groups "Western Europe" and "Iberian Peninsula" versus "Central Europe" with "Serbia" and "Southern Italy & western Alps" showed the highest differentiation. Even without the population D3, which was genetically similar to Serbia in all the analyses, a close connection of populations from central to eastern parts of central Europe and southern Europe was found. Notably, the differentiation between populations was extraordinary high in the groups from the Iberian Peninsula (61.0%) and between populations from Liguria (75.0%; data analysis of single neighbor-net regions not shown).
We found significant correlation of genetic and geographic distances with Mantel-Test over the whole dataset (r = 0.362, p = 0.001) indicating isolation by distance among populations. Isolation by distance was further present in two of the main network groups (Central Europe, Iberian Peninsula), but absent between the populations from Western Europe (data not shown).  Fig 4). A western lineage was dominant in populations from southern, western, and northern Europe, including Spain, Portugal, Andorra, France, and Belgium (Fig 3A and  3B). An eastern lineage was found in all Central European populations (Germany, Austria, Switzerland, UK, northern Italy and east of the Apennine mountain range, Czech Republic, and Serbia). High content of both lineages could be detected in populations I5 and I6 from Liguria, I1 from central Italy, D4 from western Germany and both populations from Serbia.
When both lineages were analyzed separately in Bayesian analyses, the highest probability in the western lineage of was detected for
Maximum parsimony analysis yielded 29947 equally parsimonious trees. For a 50% majority rule gene tree S. dayphyllum was manually set as an outgroup [95] (Fig 6). The trees reflected the geographic locations of the populations. The clade with Serbian populations, including

Chloroplast microsatellites
Chloroplast microsatellites revealed six haplotypes (Fig 7, Table 1), whereas we could detect four polymorphisms at locus ccmp1, three at ccmp2 and two at ccmp3. Haplotype a was present in the main group of 28 populations. Haplotype f was observed in both the Serbian

Discussion
We found high resolution and clear geographic structure in S. album populations in Europe with AFLPs, chloroplast sequence analysis and chloroplast microsatellites. A clear east/west subdivision in the investigated area, with a south-northward contact zone in western Europe was detected. AFLPs supported the results of chloroplast microsatellites and sequences (single exceptions discussed below), consistent with the findings of other studies for plants and animals [37,98].
The Iberian Peninsula populations of S. album showed high differentiation and high rarity levels. Both values suggest that populations are more isolated than those from the other regions. The microsatellites revealed distinct haplotypes of populations E4 from northern Spain and PO from Portugal. The divergences of clades from the Iberian Peninsula and western Europe seem to be ancient compared to the clades from central Europe and Italy. The Iberian Peninsula has been suggested as one of the main glacial refugia for Alnus glutinosa [19], Helianthemum sp. [13] and Rosa pendulina [7]. In all these studies, ancient haplotypes have been found on the Iberian Peninsula.
Populations AND, E2, and F5 are situated in the Pyrenees. They were clearly assigned to the Iberian Peninsula group and belonged to a distinct cluster in the Bayesian analysis of the Phylogeography of Sedum album in Europe western group (K = 7, S1B Fig). Moreover, AND and E2 had extraordinary high genetic diversity and rarity. Closed mountain valleys can harbor glacial refugia [3,27,99], and the findings of the present study support this, based on the identification of glacial refugia for S. album in the Pyrenees.
Populations from western Europe (BE, F1-F4) are closely related to populations from the Iberian Peninsula (E1-E4, PO). This northward line in western Europe represents the genetic border with populations in central Europe. Based on high genetic diversity within populations, it is possible that S. album has survived the LGM in lowland central Europe refugia, as suggested for Helianthemum nummularium [13]. Population F3 is the ancestor of the western group, according to the results of the cpDNA sequence analysis. It further showed that high rarity and diversity, which indicate high isolation in a cryptic refugium. Cryptic refugia have been shown for many plants surviving the ice ages in scattered areas with more suitable microclimate [5,6]. This was demonstrated in recent studies for several shrub and herb species in Central Europe [7,100,101]. The high amplitude of ecological niches that S. album inhabits makes survival in northern cryptic refugia possible.
Although population BE situated in north-central Europe also had a basic position in the cpDNA analysis (Fig 5 and Fig 6), the rarity of fragments was low and diversity was high, which both is contradictory to an ancestral population. A recent single dispersal event from an ancient source population would explain the ancestral cpDNA haplotype. Frequent gene flow from nearby populations via pollination Frequent gene flow via pollination by nearby populations would further explain the low rarity and high diversity levels. However, we could not detect isolation by distance within the group in Western Europe. Differentiation between the populations in general was low and rarity was moderate. Therefore, we suggest that most populations in Western Europe are exposed to frequent gene flow and constitute a contact zone for lineages from Iberia and Central Europe-east, with greater influence from Iberia.
Central European and Italian populations share one microsatellite haplotype, except for I5 and I6 from Liguria. Additionally, nine populations from Central Europe and Italy shared one sequence haplotype; further haplotypes had a close relation (except for I5 and I6 from Liguria). This group forms one derived cluster in the MP tree. In the AFLP network, populations from Central Europe and Italy could be divided in three subgroups (Central Europe-east, Central Europe-west and Italy with western Alps). Owing to poor differentiation, and low rarity and diversity, a young colonization history can be assumed for most populations from Switzerland (S1, S2) and Germany (D1, D2, D4, D5, D6).
Population D3 shares one microsatellite type with Serbian populations and is closely related to the sequence haplotype of Serbian populations. It showed high rarity, but low genetic diversity, which would be explained by a single dispersal event inducing a bottleneck and high isolation.
High rarity was also found in populations from the Czech Republic (CZ) and Slovakia (SK), along with high diversity and the most ancient cpDNA haplotype (H1) that is distributed widely in Central Europe. All analyses point towards cryptic refugia for S. album in the Czech Republic and Slovakia.
Italian populations, which are situated north and east of the Apennine (I2, I3, and I4), showed close genetic relations northward to Germany and eastward to the Czech Republic and Slovakia (S2 Fig). Therefore, gene flow or migration of S. album along the river valley of the Danube can be assumed.
Populations situated near the Alps (I7 and S2, from Italy and Switzerland, respectively) have a strong relation to populations from the middle of the Apennine peninsula. We suggest that at least these parts of the Alps are colonized by lowland southern Italian source populations after the LGM [47]. Furthermore, alpine populations have a strong connection to Central European populations and, therefore, share a colonization history with populations from the northern parts of Europe.
Mountain ranges act as contact zones for different animal and plant species [102][103][104][105]. In particular, the region Liguria is known to be a potential contact zone for lineages from different source populations of Anthyllis montana [42,43]. Notwithstanding, populations I5 and I6 showed extraordinary high rarity and differentiation, which both indicates strong isolation, contradictory to the theory of a contact zone. Survival in small-scaled bedrocks near the Alps was already assumed for different alpine species [13,30,106]. Supposedly, Liguria also harbored glacial refugia in the mountain ranges or nearby valleys for S. album.
Population I5 from north-eastern Italy (Liguria) had a strong connection to populations from Serbia. In AFLP analysis, it was also related to the group on the Iberian Peninsula. In Anthyllis montana [42], a strong connection of populations from Liguria with eastern populations-from Greece and the Balkans-has been previously shown. This connection was also present in red deer [107]. A similar picture can be drawn for S. album in the nearby population I6 in Liguria. It has a strong connection to the south-western group in Spain/Portugal, but also has some similarities with the Serbian populations. A strong relation between populations along the Mediterranean coast, from Spain and France to the middle of the Apennine Peninsula, and further eastward to Serbia was detected; this potentially reflects the northern distribution limit of S. album that first expanded west-and eastward after the LGM, before the ice and steppe regions moved further north. Moreover, dispersal along with human migrations are important considerations due to the ethnobotanical use of S. album and the longevity of seeds and shoots. Human migrations, in particular with La Hoguette culture, were documented from south-eastern Europe along the middle of the Apennine Peninsula and the Mediterranean coast [31,100], resembling the migration route of S. album.
Four populations (SRB1, SRB2, D3, and I5) constitute the most ancestral group in the MP tree. High rarity but low genetic diversity in population D3 indicates a singular long-distance dispersal event from a source in south-eastern Europe. Both Serbian populations (SRB1, SRB2) show low rarity and differentiation among populations, as revealed by the AFLP analysis. In the AFLP analysis, Central European populations were more related to Serbian populations than to populations from Iberia and Western Europe. It is possible that Serbia was colonized from source populations farther south or east on the Balkan peninsula (e.g. Greece or Bulgaria) after the LGM, and populations migrated to Central Europe. The influence of the eastern lineage on Central Europe is observed in the strong relation to population I5 and the close relation to central and northern populations, as indicated in the AFLP analysis.
Besides the clear east/west division, there was an admixture of the main AFLP lineages in populations from Germany (D4), Liguria (I5, I6), Italy (I1), and Serbia (SRB1, SRB2). Multiple colonization events would explain the observed pattern in D4 and I1. For the postulated glacial refugia in Liguria and Serbia, the combination of long-term isolation and recent frequent gene flow by pollination would explain the pattern of AFLPs and cpDNA [7]. A clear subdivision of populations from Liguria and Serbia emerged from the further subdivision of the main lineages (S1 and S2 Figs), further emphasizing the strong isolation of this area.

Conclusion
A combined analysis of the conserved chloroplast regions and AFLP markers with high resolution made it possible to infer the historical processes of S. album populations in Europe. S. album survived in glacial refugia on the Iberian and the Apennine Peninsula, and probably in eastern parts of Central Europe-as shown for several other species [3,9,108]. We found candidate areas for glacial survival in closed valleys in the Pyrenees and the mountains of Liguria, and cryptic refugia in France, the Czech Republic and Slovakia.
Recently two studies on dry grassland species demonstrated the colonization of Central Europe from source populations on the Iberian Peninsula after the LGM [100,101]. In contrast, we showed that S. album migrated to western and north-western parts of Europe from Spain/Portugal, but grasslands in Central Europe were primarily recolonized by cryptic source populations or source populations coming from the Apennine Peninsula and eastern Central or south-eastern Europe.