Out of 15 mtDNA regions surveyed (Table 1), intraspecific polymorphism was detected for only two loci in Scots pine. Polymorphism was detected de novo for the intron 1 of nad7 with three length variants. They were 1175, 1170 and 1143 bp long and are hereafter called haplotypes A, B, and C, respectively. The size differences were caused by two single indels: one of 32 bp exclusive to haplotype C and one of 5 bp exclusive to haplotype B (Table 2). As previously reported from a more limited sampling 6, two size variants were detected for the intron B/C of nad1 among the 986 samples analysed herein. These fragments differed by a 31 bp indel and are hereafter called haplotypes A (217 bp) and B (248 bp). A third variant restricted to Italy was recently reported for this gene 30. It was not detected in our sampled trees which were all from outside the Italian Peninsula. Primers for these two mtDNA regions were also tested on 100 to 500 individuals for each of five other pines of the subgenus Diploxylon (P. resinosa, P. mugo, P nigra, P. heldreichii, and P. contorta) and positive amplifications were obtained (data not shown). The first four species belong to the section Pinus, but the first three are members of subsection Pinus (as for Scots pine) and Pinus heldreichii is member of subsection Pinaster. As for P. contorta, it belongs to section Trifoliae, subsection Contortae. Fragment length polymorphism was detected for both nad1 intron B/C and nad7 intron 1 in P. heldreichii. For P. nigra, only nad1 intron B/C showed polymorphism. No polymorphism was detected in P. mugo and P. resinosa. For P. contorta, polymorphism was only observed for nad7 intron 1 involving the same minisatellite-like marker as previously reported for this species and the closely related P. banksiana 20.

For Scots pine, the nomenclature used for mitochondrial multi-locus haplotypes (mitotypes) followed a two-letter code after the haplotypes observed for nad7 intron 1 (A, B, C) and nad1 intron B/C (A, B), respectively (Table 2). Among the four observed mitotypes, the most abundant one was AA (72.3 %), followed by BA (17.3 %), CA (5.8 %), and AB (4.6 %) (Table 3). The minimum spanning tree indicates that the mitotypes were more or less arranged as a star phylogeny with little sequence variation among them and no recombination (Figure 1B). The observed number of mitotypes per population (nh) was 1 or 2 without exception, with an average of 1.50 (Table 3). The mitotype diversity index per population (H) was low to moderate (average of 0.141) and ranged from 0 to 0.500 (Table 3). Population differentiation was high among the 54 populations, with G_ST and N_ST values of 0.657 and 0.685, respectively (Table 4). When excluding the six more diverse populations from a presumed zone of contact in central and northeastern Europe between two distinct mtDNA lineages (see below), the G_ST and N_ST values increased to 0.746 and 0.767, respectively, among the remaining populations. However, the presence of a formal phylogeographic structure was not supported, as N_ST was not significantly higher than G_ST. Lack of test power is likely the reason, given the small genetic divergence observed among the four mtDNA variants detected.

The geographic distribution of mitotypes appeared to be essentially non-random, with clusters of mitotypes geographically quite well delineated (Figure 1A, see also SAMOVA test below). The Mediterranean region showed the highest diversity with three of the four mitotypes revealed in the study. The cosmopolitan haplotype AA was largely distributed and found in most of the populations sampled. Geographic structuring was evident for other mitotypes. Mitotype CA was present only in Asia Minor, in the Pontide Mountains of Turkey (populations #14 to #16). Mitotype AB was found fixed in some high mountain populations of the Iberian Peninsula (Iberian cordillera – Zaragoza/Valencia provinces) (#52 and #54) and was observed as a rare variant in some populations of the Balkans (#3, #5, #6, #8 and #10). The distribution of mitotype BA was quite unexpected: it was largely distributed in the lowland region of middle to northern latitudes in eastern Europe. It was prevalent in the Baltic region and quite frequent in Russia, west of the Ural Mountains. But it was also detected in Siberia (population #44). A large zone composed of populations with variable mixed genetic background (mitotypes AA and BA) was observed in central and northeastern Europe. It includes most if not all populations bearing mitotype BA. The zone extends from the southeast of Russia to the Baltic States (see populations #32, #33, #35, #38, #40, #41, and #43). The within-population mitotype diversity of these populations (H = 0.463) was much higher than elsewhere.

SAMOVA was conducted repeatedly by increasing the value of the number of population groups (K) until the F_CT value reached a maximum of 0.785, which was obtained for K = 4 (Table 4). For K = 3, a maximum F_CT value of 0.720 was obtained. The four groups confirmed trends from visual inspection and corresponded to: (I) the majority of populations from Europe and Asia, (II) northeastern Europe, (III) Asia Minor and (IV) the high mountains of the Iberian Peninsula. Group I was the most widespread geographically, including 44 populations (Table 4). It contained mostly the cosmopolitan mitotype AA and, at a lower frequency, mitotype BA and AB. Differentiation among the populations of the group was moderate. Population differentiation within each of the three other groups (II, III, and IV) was smaller. Group II from northeastern Europe was largely dominated by mitotype BA. The two remaining groups III and IV were highly homogenous. Group III regrouped populations from Turkey fixed or nearly fixed for mitotype CA, and group IV regrouped populations from the Iberian Peninsula fixed for mitotype AB.

DNA sequence variation is low in the mtDNA of conifers [e.g. 4750]. Accordingly, little intraspecific variation has been found in the mtDNA of P. sylvestris in western Europe, including four RFLP variants for the cox1 region 5 and three size variants at the locus nad1 intron B/C, of which two were found endemic to the Iberian and Italian Peninsulas 630. The three new size variants reported herein for the locus nad7 intron 1 indicate new region-specific variants for northeastern Europe and Asia Minor. No other length polymorphisms or substitutions were detected in Scots pine in spite of the screening of 15 distinct mtDNA regions. Highly informative minisatellite-like motifs such as those rarely detected in some spruce 5253 and pine taxa 20 could not be detected in Scots pine.

The overall level of population differentiation estimated in this study from mitotype variation at the nad1 intron B/C and nad7 intron 1 loci was rather high and similar to that reported in earlier studies restricted to Scots pine populations from western Europe based on RFLPs of the cox1 region (G_ST = 0.83; 5) or indels of the nad1 intron B/C (G_ST = 0.60; 6). Such estimates are in line with averages reported for maternally inherited markers for a number of conifer and angiosperm taxa 54. The amount of population differentiation determined by nuclear markers such as isozymes was generally much smaller in Scots pine with G_ST values ranging between 0.025 and 0.076 4355565758596061]. However, these studies were carried out mostly at the regional level. Among-population differentiation was also much higher for mtDNA markers than for paternally inherited cpDNA markers in a Scots pine study covering much of the species range (G_ST = 0.11; 62), as usually observed in other conifers [e.g. 5463].

The present study confirms the existence of a genetically distinct set of populations in the high mountains of the Iberian Peninsula 56 while variants discovered in the intron 1 of nad7 signal distinctive evolution in Asia Minor and northeastern Europe. Haplotype B at locus nad1 intron B/C was observed mostly in the Iberian Peninsula in some high elevation populations of the Zaragoza/Valencia regions where it was fixed or nearly fixed. It was observed only sparingly west of Catalonia. Such a distribution indicates that these Catalan populations have been isolated during glaciations and have not contributed to the recolonization of Europe [56 and 30]. A distinct dynamics of Catalan populations from the rest of the Iberian Peninsula has also been reported for other species like Aleppo pine 64, Maritime pine 965, and white oaks 8. In the present study, haplotype B of nad1 intron B/C has also been discovered, albeit at low frequency, in five other populations located in Bulgaria, in the Balkan Peninsula, thus far away from the center of prevalence for this haplotype in the Iberian Peninsula. Long-distance seed dispersal could be invoked but is unlikely over such long distances, given the heterogenous topography and relative isolation of Iberia. The probability of such long-distance movements would be higher in the northern and more eastern parts of Scots pine natural range, given the lack of major physical barriers. Another perhaps more plausible explanation for the rare presence of haplotype B of nad1 intron B/C in some Balkan populations could be that it is ancient and was present in more than one southern ice refugia. Its high or low frequency, depending on the region considered, could be attributable to stochastic variation, especially if southern refugia populations suffered from severe bottlenecks. The weak occurrence of this haplotype in the Balkans might also result from human activity, as previously suggested for the rare presence of this variant in Poland 6. Whereas the hypothesis of seed source transfer cannot be totally ruled out, it is difficult to sustain because of (i) the lack of hard evidence for such seed source transfer between the Balkans and the Iberian Peninsula in the past; (ii) natural stands in which seeds were collected existed long before forest management activities began in Europe; and (iii) the relative conservatism of forest seed transfer guidelines in the Balkan region 66. A third explanation could relate to ancient mtDNA capture from hybridizing species in the region 6768. Accordingly, phylogeographical studies of other pine species in the region using the same gene locus might help clarify the issue.

The detection and high prevalence of haplotype C of nad7 intron 1 in Asia Minor populations only is a strong signal for the presence of a genetically distinct glacial population. This relictual population would have experienced severe bottlenecks, because populations were all fixed for this haplotype except for one tree harbouring the cosmopolitan haplotype A. After glaciation, it would have remained isolated, eastward by the Caucasus Mountains, northward by the Black Sea, and westward by the Bosphorus passage. The cold and arid climate in this region during the Holocene might have also impeded the expansion of this population 69, but not its survival as Scots pine can survive under very harsh conditions 70.

The high prevalence of haplotype B of nad7 intron 1 in northeastern Europe suggests the presence of a genetically distinct glacial population at more northern latitudes than previously thought, an idea that had already been suggested for Scots pine 5. A recent mutation during the Holocene giving rise to haplotype B of nad7 intron 1 is unlikely, given the widespread distribution of this haplotype. As well, homoplasy is unlikely, given the low rate of variation of angiosperm and conifer mitochondrial introns 4749. A large number of populations had mixed composition of both haplotypes A and B of nad7 intron 1 in central and northeastern Europe. Their diversity index was highest among all populations surveyed, which could result for the meeting of two genetically distinct glacial lineages (A and B), or indicates that the glacial population harboring haplotype B had mixed genetic composition. The uneven distribution of haplotype B of nad7 intron 1 among these populations, with a higher prevalence of haplotype B as one moves northward, could result from successive founder events during colonization 71, as recently proposed for European hedgehogs 72. This is likely, given than that migration from more southern latitudes was necessary to colonize these previously glaciated areas.

There is some evidence from fossil and palynological data that cold adapted tree species occupied mid-altitude sites in the mountains of southern Europe during the last glacial stage 13. Moreover, the southern peninsulas of Iberia, Italy, the Balkans, and Greece, which are separated from the rest of Europe by major mountain ranges such as the Pyrenees, Alps and Caucasus, were identified as major glacial refugia for many species [1430 and 65]. Our genetic data indicates that a genetically distinct glacial refuge existed in Asia Minor as well. Tree pollen, including that of the genus Pinus, was recorded in the Balkans about 14 000 years BP 13 while it was barely present as late as 10 500 years BP in southwestern Turkey where the climate was cold and dry with steppe vegetation 69. Pine pollen attained its maximum only in early Holocene in western Anatolia and in mid-Holocene in more eastern areas, likely because of similar climatic limitations 7374. Thus, it is apparent that tree populations would have been scarce in the region at LGM, 21 Kyr ago. The refuge is also unlikely to have contributed much to the postglacial colonization of Europe, given the genetic discontinuity detected in Scots pine. The peculiar location of Turkey between the Black and Mediterranean seas would have promoted isolation from more northern areas. It is likely that other elements of the steppe vegetation in the region also represent relicts from the pre-Holocene era and suffered similar effects during glacial periods, as shown recently for field mouse 75. As such, it is expected that they would show a distinct genetic background, as that detected in Scots pine.

The exact location and size of the Scots pine glacial population giving rise to the large spread of haplotype B of nad7 intron 1 in eastern or northeastern Europe remain unknown. Based on fossil data, Scots pine glacial populations existed in the Balkans and possibly even in central Europe [Litynska-Zajac M. cited by Stewart and Lister 2223]. The large spread of haplotype A of nad7 intron 1 throughout Europe is likely the consequence of an expansion from glacial populations from that region 30. Such spread northwestwards towards Germany and France and northeastwards throughout Eastern Europe and Siberia correlates geographic dispersion patterns and colonization routes inferred for a grasshopper 76, for European beech 77, for the oaks 4, and for Norway spruce 5278. However, some of these colonization routes are still open to alternative interpretation. Furthermore, such observations do not correlate well with the exceptionally northern location of haplotype B of nad7 intron 1.

The peculiar northern distribution of haplotype B of nad7 intron 1 indicates that Scots pine has likely survived in a refuge distinct from those usually recognized in the southern peninsulas of Europe, perhaps under the form of scattered glacial populations spread over sizeable latitudinal areas reaching European mid-latitudes not far from the ice front. Scots pine is known to survive and grow reasonably well upon permafrost 72. Evidence is also accumulating regarding the existence of glacial populations located at more northern latitudes [e.g. 2528 and 29] and their significant contribution to Holocene colonization 2225. In addition, if the mixed zone observed in north-central and northeastern Europe between the cosmopolitan AA mitotype and the northern BA mitotype represents truly a zone of secondary contact, the northern location and shape of this zone would not correspond well to the meeting of two ancestral lineages both located in or near southern European peninsulas. While the existence of such zones with typically higher genetic diversity has been reported for other tree species in Europe [34 and 16] as well as in North America 1920 where ancestral lineages from southern latitudes would converge at more northern latitudes, this possible zone of contact would more likely correspond to the junction between a lineage moving northwards from the Balkans or from even more east (the AA lineage), and a lineage of more northern or northeastern origin (the BA lineage). In one rare occurrence of similar observation, Taberlet et al 3 recognized a contact zone of migration fronts as far north as in Scandinavia for Picea abies. They proposed that the area was presumably colonized from the south and from the northeast by populations originating from different refugia.

While the current areas of maximum abundance of haplotype B of nad7 intron1, Finland and northwestern Russia, were completely glaciated at LGM 79, it is likely that trees colonizing the region would have migrated northward from a region anywhere west of the Ural Mountains. There are indications from general circulation models that the climate during the last glacial maximum was warm enough to allow Scots pine to survive well as far north as above 50°N in isolated regions of northern Ukraine and southwestern Russia, for instance 30. A location more south is also possible if founder events after long-distance dispersal are hypothetized, followed by drift and lineage sorting to account for the high abundance of haplotype B of nad7 intron 1 as one moves into Finland and northwestern Russia. However, this hypothesis becomes unlikely for long-distance dispersal from refuges located in the southern European peninsulas, given the absence of haplotype B of nad7 intron 1 in southern populations of Scots pine.

More mtDNA polymorphisms will need to be discovered in order to distinguish Scots pine disjunct populations from eastern Europe and Asia and investigate if significant glacial vicariance factors existed in this vast region. Siberian populations where the cosmopolitan mitotype AA was dominant might have been established from a glacial population located in southwestern Siberia, in the upper Irtysh River Valley 808182, or from a more southern location. As for Scots pine populations located in the far-east in China where the cosmopolitan mitotype AA was found fixed, it is not clear if it is derived from a more western refuge located in southwestern Siberia (see above) or if a refuge existed in northeastern China. Palynological evidence suggests that pine was present at several locations in the region during the last glaciation, but it is unclear if the fossil pollen was representative of Scots pine as many pine species are endemic to the region 82.

Fifty-four natural populations were sampled and their location was chosen in order to represent most of the natural range of the species in Europe and Asia. Twenty populations were from the eastern part of the Ural Mountains and Asia (Russia and China) as well as from Asia Minor (Turkey), and the remaining 34 populations from 13 European countries (Figure 1 and Table 3). Each seed lot represented a bulk of seeds from a minimum of 50 individuals with a minimum distance of 100 m between them. Seed samples were collected from each population and stored in darkness at 4°C until they were processed.

Seeds from each population were soaked on moistened filter paper in Petri dishes at 26°C under a photoperiod of 14 hours for 2 days until DNA extraction. DNA was extracted from the haploid megagametophyte for each of 9 to 20 seeds per population using a genomic DNA mini preparation kit (Sigma-Aldrich), for a total of 986 DNA samples and an average of 18.3 samples per population (Table 3). A screening for mtDNA polymorphism was conducted by assembling an exploratory panel of 26 out of the 986 DNA samples, each representative of a different population and ensuring that geographically widespread populations were represented.

A total of 15 mtDNA regions were screened for polymorphisms using various techniques (Table 1). Polymorphism was detected for only two regions, nad1 intron B/C (2 haplotypes), and nad7 intron 1 (3 haplotypes). DNA samples bearing different haplotypes for each of the variable mitochondrial regions were selected (3 DNA samples per locus and per haplotype from different populations) and sequenced in order to determine the exact nature of every fragment length polymorphism and to detect the presence of potential fragment length homoplasies. Direct sequencing of the two DNA strands was carried out on an automated 3730XL DNA analyser (Applied Biosystems) with the dideoxynucleotide chain termination procedure using the appropriate amplification primers and Big Dye Terminator kit-V.3.1 (Applied Biosystems). The complete sequences for nad7 intron 1 haplotypes have been deposited in GenBank (GenBank: DQ665913 to DQ665915). The complete sequence for nad1 intron B/C was already available in GenBank 6. Multi-locus mtDNA haplotypes (mitotypes) were then defined by considering single-locus mtDNA genotypes simultaneously and their relationships were determined by statistical parsimony analysis of the aligned sequences using the program TCS 86.

Single-locus haplotypes were determined for each of 986 individuals with the internal primers pair nad1H and nad1I for nad1 intron B/C, as described by Soranzo et al 6, and the primers nad7 intron1 forward and nad7 intron1 reverse for nad7 intron 1, as described by Jaramillo-Correa et al 19. Primers were synthetized by Invitrogen. DNA was amplified in a PTC225 thermal cycler (MJ Research) using Platinum Taq DNA Polymerase (Invitrogen) in a 12 μl reaction volume, following protocols published by Soranzo et al 6 for nad1 intron B/C and Jaramillo-Correa et al 19 for nad7 intron 1. PCR products of nad7 intron 1 were digested with the restriction enzyme Sau3A I (Promega) and then separated through 8 % non denaturating polyacrylamide gels (in TBE) in order to detect putative cleaved amplified polymorphic sites (CAPS). PCR products of nad1 intron B/C were electrophoresed through 2 % agarose gel.

The number of observed mitotypes per population (nh) and the total mtDNA diversity per population (H; equivalent to the expected heterozygosity, H_E, for diploid data; 87) were estimated. Population differentiation was estimated using two fixation indices, G_ST and N_ST, using PERMUT 88. The first statistics is calculated based solely on mitotype frequencies, whereas the second one takes into account the genetic relatedness among mitotypes. Thus, if the estimated N_ST value is higher than the G_ST value, it indicates that closely related mitotypes tend to cluster in the same area, supporting the presence of a formal phylogeographic structure.

Geographic population structure was assessed using a spatial analysis of molecular variance SAMOVA 1.0 8990. A matrix of pairwise distances between populations was first constructed using mitotype frequencies, as well as a matrix of pairwise geographic distances between populations. The method implemented in SAMOVA employs a simulated annealing procedure and uses allele frequency data along with geographic coordinates of the sampled populations to identify population groups maximizing genetic differentiation. We determined the most likely number of K groups by repeatedly running SAMOVA with variable numbers of groups and by choosing the number resulting in a maximum F_CT value 90. The configuration with the largest F_CT value among the 100 tested was retained as the best grouping of populations.