Pleistocene glacial cycles have been recognized as one of the major drivers of polyploid speciation in several alpine/arctic genera, including Cerastium
Saxifraga and Vaccinium [19; as reviewed in . Polyploidization involving differentiated progenitors might have occurred in glacial refugia, where several species survived glacial maxima (e.g., Androsace brigantiaca ), or via secondary contact between populations that became isolated during glacial maxima and reconnected during interglacials [16, 17], as proposed for Primula sect. Aleuritia .
The Pleistocene time frame for the evolution of Primula sect. Auricula inferred from both a section-wide ITS phylogeny  and a Primula-wide cpDNA phylogeny [De Vos & Conti, unpublished results] suggests that glacial cycles likely influenced the origin of P. marginata. In the present study, we use multiple lines of evidence to reconstruct the likely evolutionary history of P. marginata in relation to Pleistocene climate-dynamics. Our combined results allow us to propose possible superimposed processes that left distinctive signatures on the distribution of cytotypes and phylogenetic structure of the P. marginata chloroplast and nuclear genomes, suggesting events that might have occurred at different points in the evolution of the lineage. The complementarity of information between cytotype variation and nuclear and organellar genomes has proven indispensible to elucidate complex evolutionary histories in other boreal taxa, e.g. Vaccinium , Artemisia , and Nymphaea .
Flow cytometric analyses (Table 1) allowed us to expand the relatively limited, available information on cytotype variation (e.g., 41 individuals from 11 localities; [31, 33, 34]) to a survey of 100 individuals from 17 populations of P. marginata. Failure to detect any intermediate cytotypes nor any variation of ploidy levels within populations (Table 1) are congruent with the preliminary results of crossing experiments suggesting reproductive incompatibilities between the two cytotypes [Minuto et al., unpublished results]. Our investigations also enable us to confirm that the hexaploid and dodecaploid populations of P. marginata occur primarily in the western and eastern parts, respectively, of the species’ distributional range (Figure 2A).
Both adaptive and non-adaptive processes can explain the infra-population uniformity of cytotypes and their geographic separation. In the adaptive scenario, novel genetic combinations in the polyploids may allow them to adapt to different environmental conditions corresponding to specific geographic areas [7, 13]. In the non-adaptive scenario, hybridization between cytotypes is thought to be either non-viable or produce plants with lower fitness, gradually leading to the elimination of the minority cytotype through frequency-dependent mechanisms (‘minority cytotype exclusion model’, ). These non-adaptive processes, while producing cytological uniformity, promote differentiation in a stochastic manner that does not usually produce distinct morphological and ecological characteristics [43, 44]. In P. marginata the lack of morphological distinctiveness of the two cytotypes  and the absence of any obvious new ecological preferences in the dodecaploids  all fit the predictions of the non-adaptive scenario.
The two cytotypes of P. marginata do not form distinct clades either in the cpDNA or in the nrDNA phylogenies (Figures 2B, Figure 3), a pattern that seems more congruent with the stochastic nature of non-adaptive processes than with a scenario implying divergence between cytotypes adapted to distinct ecological conditions in different parts of the species range [13, 42]. Despite the lack of reciprocal monophyly between the western, hexaploid and the eastern, dodecaploid populations, respectively, a signal of some geographic structure is detectable in the cpDNA topology. In this tree, the sequences of five hexaploid populations from the western part of the P. marginata range form a clade with a population of P. latifolia located further to the East (lat2; Figure 2B, clade 2), and a mixture of sequences from hexaploid and dodecaploid P. marginata populations eastward of the previously mentioned ones form a clade with two other samples of P. latifolia (Figure 2B, clade 1). The sharing of cpDNA haplotypes between P. marginata and P. latifolia might be explained by reticulation or incomplete lineage sorting from a polymorphic ancestor [45, 46]. These processes are not necessarily mutually exclusive and are difficult to disentangle. However, assuming that the DNA sequences under exam are not under selection, a geographical pattern, as observed in our case, more likely results from introgression than lineage sorting .
In light of the non-adaptive scenario invoked above to explain the spatial separation between hexaploid and dodecaploid populations of P. marginata and the Pleistocene time frame for the evolution of Primula sect. Auricula [29; deVos, unpublished results], it is plausible to propose that abiotic processes, likely involving range fragmentation and/or shift during glacial maxima, contributed to the geographic signature in the cpDNA tree. More specifically, the clustering of all sampled dodecaploid populations of P. marginata with hexaploid populations from the southern portion of its distribution (Figure 2B clade 1), located in the glacial refugium of the Maritime Alps [48–50], might indicate that refugial populations from this area likely played a role in the origin of the dodecaploids. This explanation of geographic structure in the cpDNA phylogeny is consistent with the general interpretation of refugia on the southern side of the Alps as reservoirs of evolutionary potential during the climatic oscillations of the Pleistocene [29, 38, 48, 50–54]. To summarize, the cpDNA tree indicates that, among the species of sect. Auricula that co-occur in the western Alps and are known to form hybrids with P. marginata either in nature or in cultivation [25, 30], the latter shares a more recent common ancestor with P. latifolia than with P. allionii or P. hirsuta, while it excludes a close relationship with the allopatric P. auricula .
The nrDNA tree does not corroborate the common ancestry between P. marginata and P. latifolia supported by the cpDNA topology. The greater partitioning of ITS variation according to taxonomic groups (21.39% and 24.01%) than to the clades supported by the cpDNA phylogeny (4.33%; Table 3) provides a further indication of the topological conflict between the two phylogenies. In the nuclear phylogeny, all sequences from three populations of P. latifolia are included in a monophyletic group, while those of P. marginata and P. allionii are unresolved. Some sequences from a southern population of P. allionii (all2) share more recent common ancestors with sequences from hexaploid or dodecaploid populations of P. marginata (clades III, IV; Fig. 3) than with sequences from a northern population of P. allionii (all1), which form a well-supported clade (clade II; Figure 3). The non-monophyly of ribotypes from the same species may result from introgression between P. marginata-like and P. allionii-like ancestors, eventually leading to the homogenization of all2 sequences towards the P. marginata ribotypes. The interdigitation of nrDNA sequences in the phylogeny is reflected in the lack of distinct species clusters in the corresponding PCoA scatterplot (Figure 4). Congruently, most ITS variation is allocated within individuals (62.30-69.91%) rather than between groups in the AMOVA (Table 3). Our results also contradict the sister relationship between P. marginata and P. latifolia supported by one of 7770 equally parsimonious trees derived from direct sequencing of ITS amplicons in a previous study, where no cloning was performed . In summary, nuclear ribotypes from different individuals of a P. allionii population located in the southern refugium of the Roya Valley in the Maritime Alps  and some hexaploid and dodecaploid populations of P. marginata coalesced more recently than haplotypes within the respective species, indicating a close and complex evolutionary history for their nuclear genomes.
Discrepancies between the phylogenetic signal of chloroplast and nuclear genomes have been found in several other polyploid complexes of the Northern hemisphere (e.g., Vaccinium uliginosum , Primula sect. Aleuritia , Cerastium , Viola ; see review in ). Explanations for cytonuclear conflicts range from differential substitution rates between the two genomes, hybridization/introgression, paralogy and incomplete homogenization or lineage sorting, especially of the nuclear sequences, or a combination of various processes [57–61]. Reciprocal illumination between different lines of evidence may help to favour some explanations over others. For instance, the neutral theory of molecular evolution  predicts that nuclear sequences should take longer to coalesce than organellar sequences, due to the larger, genetically effective population size of the former, especially in polyploid species [63, 64]. Conversely, concerted evolution is supposed to speed the homogenization of multiple copies, as in the nrDNA region [59, 65, 66]. Another observation relevant to the interpretation of discrepancies between cpDNA and nrDNA phylogenies is that cpDNA sequences tend to evolve more slowly than ITS sequences, thus the former might be more suitable to track deeper evolutionary events, while the latter might better capture the phylogenetic signature of more recent events (e.g., [39, 67]). Therefore, our phylogenetic results, taken together, seem compatible with the hypothesis of an initial homoploid introgression of the chloroplast genome between a P. latifolia-like ancestor and a P. marginata lineage, resulting in the persistence of cpDNA sequences that are more closely related to heterospecific than conspecific sequences (Figure 2B). The initial episode of chloroplast capture was probably followed by the separation between western and eastern populations of P. marginata, possibly driven by advancement of glaciers during glacial maxima, and the subsequent origin of the dodecaploids (Figure 2A). The occurrence of ITS clones from the all2 population of P. allionii in clades with dodecaploid and hexaploid individuals of P. marginata suggests that an ancestral southern population located in the Roya Valley, which was ice free in the Last Glacial Maximum , might have played a crucial role in the origin of P. marginata dodecaploids (Figure 2A, Figure 3). However, the lack of intermediate cytotypes in P. marginata (Table 1) and of any admixed individuals in an AFLP survey of P. marginata and P. latifolia  implies that the three species might have been evolutionarily isolated more recently, an interpretation also proposed by Kadereit et al. .
The higher proportion of ITS additive polymorphisms in dodecaploid than in hexaploid accessions of P. marginata (71% vs. 19.3%, respectively; Table 2) and higher total nucleotide diversity of ITS clones in the sequences of dodecaploid vs. hexaploid individuals (0.032 vs. 0.015) are congruent with an allopolyploid explanation for the origin of the dodecaploid populations of this primrose . At the same time, the occurrence of shared additivity between P. allionii and P. marginata suggests a role of the former in the evolutionary history of the latter. The allopolyploid interpretation for the origin of P. marginata dodecaploids might appear to conflict with their lack of morphological differentiation [25,34; Casazza, unpublished observations]. Our case study might represent an example where the allopolyploids resemble one of the putative parental lineages, a situation that has been found also in Mimulus  and Centaurea toletana ; as reviewed in  and is compatible with the non-adaptive scenario [43, 44] proposed above to explain the geographic separation between hexaploid and dodecaploid populations.