DNA sequence divergence clearly plays a leading role in shaping the phenotypic variation observed between most taxa e.g. [1, 2] but does not explain all forms of adaptive phenotypic differentiation [3–5]. Epigenetic modifications of DNA and histones, the core components of chromatin are able to influence the expression of the underlying genes and so phenotype [6]. Of these epigenetic mechanisms, the best studied and the one to show more examples of transgenerational stability is cytosine DNA methylation [6, 7]. Evidence has also been found suggesting that heritable phenotypic variation observed in natural populations can be due to stable epigenetic variation, even in the absence of genetic variation and could play a role in plant adaptation and evolution [5, 8–10]. The scope for epigenetic mechanisms driving morphological differentiation is perhaps best illustrated when genetic variation is lacking. Understanding the relative importance of genetic and epigenetic sources of phenotypic variation where both systems are operating is therefore attracting increasing interest. For example, Laguncularia racemosa is a mangrove plant species that shows low genetic variability but in populations from distinct, nearby habitats, cytosine methylation variation among individuals correlates more closely with environmental variation than does genetic variation [11]. In the perennial Viola cazorlensis, cytosine methylation patterns were found to be partitioned and positively correlated with adaptive genetic variation [12]. Also, in populations of individuals with reduced or negligible genetic variation such as those of triploid asexual dandelion lineages (apomixis; diplospory), changes in genomic methylation patterns are found between individuals [13].

The genus Limonium Miller (sea-lavenders; Plumbaginaceae) has long been recognized to have a history of recurrent hybridization and polyploidization, and comprises 150 [14] to 350 taxa recognized across coastal, steppe and desert regions (e.g. [15, 16]). This wide range is due to the description of new taxa, mainly microspecies from geographically restricted areas. In this genus, a sporophytic self-incompatibility system is linked with pollen-stigma dimorphisms, A-pollen type grains germinate on papillose stigmas and B-pollen type germinate in cob-like stigmas, while the complementary combinations produce no successful fertilization [17, 18]. Most sexual species of Limonium usually have a dimorphic self-incompatibility system (both pollen and stigmas are dimorphic) while agamospermous species are generally monomorphic and have monomorphic populations [14, 19]. Determination of these characters in individuals from natural and/or experimental populations has since long been used as an indirect method for estimation of each species reproduction mode [14, 17, 18]. The high number of polyploid taxa has been explained to be a consequence of this self-incompatibility system and the ability of polyploid hybrids to produce seeds asexually via apomixis (agamospermy; asexual seed formation) [14, 17–20]. Ixeris-type embryo sacs with non-haploid eggs are found in triploid (2n = 3x = 27) Statice oleaefolia var. confusa[20]. In triploid and tetraploid agamospermous species of the L. binervosum (G. E. Sm.) Salmon group diplospory followed by parthenogenesis is reported [21, 22]. Molecular phylogenetic studies have tried to resolve the taxonomic complexity within this genus in a global perspective using nuclear DNA sequence information [23] and plastid DNA [24–26].

In Continental Portugal about 15 Limonium species have been recognized with ecological importance for plant communities of the Atlantic and Mediterranean coastlines [15, 27]. Among these, the L. ovalifolium complex consists a group of three sexual diploids (2n = 2x =16): L. ovalifolium (Poir.) O. Kuntze, L. nydeggeri Erben and L. lanceolatum (Hoffmanns & Link) Franco [28]. The first species has a broader distribution including several Sites of Community Importance (SCI) for the Mediterranean biogeographical regions [29] in the West (Estremadura), Southwest Alentejo and Algarve coastlines [15, 30]. Conversely, the Lusitania endemic L. nydeggeri and L. lanceolatum have more restricted distributions; the former is restricted to West and Southwest Atlantic sea-cliffs whereas the latter is found in the Southwest and South coastlines [28, 30]. Limonium tetraploid taxa include among others, the Lusitania endemic apomict, L. multiflorum Erben [14, 15] which exhibits both tetraploid and aneuploid tetraploid cytotypes (2n = 4x = 35 - [14]; 2n = 4x = 32, 34, 35, 36 – [31]) and the aneuploid tetraploid apomict L. dodartii (Girad) O. Kuntze (2n = 4x = 35) which most frequently grow on maritime cliffs in the province of Estremadura. A third tetraploid, L. vulgare (2n = 4x = 36), a sexual species [19, 32, 33], grows in salt marshes [15, 30].

Dominant genetic markers, such as those generated by Amplified Fragment Length Polymorphism (AFLP), are valuable for assessing genetic diversity within and between populations [34] and for inferring taxon differentiation [35], especially in species for which codominant markers are unavailable. Some recent publications have added data on natural epigenetic variation in animal and plant species by sampling cytosine methylation using Methylation-Sensitive Amplification Polymorphism (MSAP) technology [36–38], a modification to the original AFLP protocol that compares product profiles generated by methylation-sensitive/insensitive isoschizomers. Central to the technique is the differential behavior of the two isoschizomer restriction enzymes (HpaII and MspI) in the presence of cytosine methylation in the CCGG context. HpaII is inactive if one or both cytosines are methylated at both DNA strands, but cleaves when one or both cytosines are methylated in only one strand. MspI, by contrast, cleaves C5mCGG but not 5mCCGG. Comparison of the profiles generated by each enzyme from each individual allows the assessment of the methylation state of the restriction sites and so provides a relative comparison of genetic and epigenetic variability. Several reports suggest that only methylation marks in the CG context can be transmitted between generations and so have potential for stable adaptive significance [39]. Furthermore, recent data shows that although non-CG methylation can be inherited, only inherited CG methylation is inversely correlated with gene expression [40]. Since only HpaII is affected by methylation of this kind, for simplicity, in this study we refer to profiles from this enzyme as epigenetic (meaning potentially transgenerationally stable and epigenetic) whereas MspI is insensitive in this sense and so can only detect transgenerationally relevant genetic variation. The focus of this study was to therefore to use MSAP analysis as the primary tool in comparing the extent to which genetic and epigenetic diversity in natural populations of diploid and tetraploid Limonium species correlate with species identity and ploidy.

The effects of hybridization, polyploidy and apomixis have all combined to shape radiation currently seen in Limonium species [26, 31, 43]. This evolutionary model has been used to explain multiple series of complex aggregates of sexual diploid species and asexual polyploid hybrids which are perpetuated through gametophytic apomixis [44–48]. In other plant groups, hybridization and polyploidy have combined to generate genetic and phenotypic complexity in the form of classical polyploidy pillar complexes. In these situations species delimitations typically become blurred at the higher ploidy levels and interploidy discrimination can also become difficult for some taxa. The many examples of this type of species complex include the Festuca ovina aggregate [49], Dactylis glomerata[50] and Knautia arvensis species groups [51]. The additional and intermittent appearance of facultative or obligatory apomixis, as seen in triploid and tetraploid Limonium species [20, 21], adds another layer of complexity for species delimitation and diagnosis. In these instances the phenotypic range of taxa can be highly variable, as can their morphological distinctiveness and stability (e.g. [52]). The cumulative effect of these processes is typically manifest in the recognition of a large number of microspecies, with the diagnosis and genetic characterisation of many of the resultant taxa often being highly demanding (e.g. Limonium, reviewed in [26]). The publication of a comprehensive revision of Limonium species in Southwest Europe by Erben [14] in which 59 species (13 new to science) were fully described allowed for several taxometric studies of geographically related species in this genus to be conducted. For example, various authors used a taxometric approach to study several sexual and agamospermous species of the L. binervosum (G. E. Sm.) C. E. Salmon complex from Western Europe [53], and also to examine the closely related L. vulgare and L. humile species in the British Isles [33]. Morphometric studies have also been used at population level. For instance, some authors studied the taxometric relationships between triploid or aneuploid tetraploid L. binervosum agamospermous colonies in the British Isles [42, 53]. A similar strategy is applied to describe morphological differentiation patterns between individuals of a L. dufourii population from Eastern Spain [54].

In the current study, CDA was applied using ten morphometric traits collected from representative individuals of three diploid and three tetraploid Limonium species from Portugal. As previous studies had suggested [14, 33, 41, 42], the collective use of these characters in CDA was sufficient to provide clear morphological differentiation between species at each ploidy level. This differentiation is based primarily on the use of seven morphometric variables, viz: MOBL, MCL, MMBL and MIBL for the first axis and MMBW, MOBW and MIBW on the second. Overall, these analyses not only provided clear separation of all diploid species but also indicated that L. lanceolatum and L. ovalifolium share a closer phenotypic affinity. Separation of the tetraploid species was also possible but with intermediate individuals blurring the boundaries between L. vulgare and L. dodartii, and between the latter and L. multiflorum. The rather surprising finding, however, lay in the clear phenotypic separation of tetraploid species from the diploids on the basis of the bract and calyx characteristics MCL, MOBL, MMBL and MIBL rather than plant size features more usually associated with ploidy level changes (e.g. see [55]). The relative importance of genetic and epigenetic processes in shaping the observed phenotype structuring among species and ploidy levels in Limonium has thus far remained elusive.

Previous works on the analysis of genetic variation and population genetic structure in other Limonium species have deployed a wide range of molecular marker systems to infer the importance of genetic structuring in defining interspecies delimitation. There has, nevertheless, been considerable evidence that modest but significant levels of genetic variation does occur within species in the genus. For example, studies on the presumed agamospermous triploid L. dufourii have invariably revealed low but substantial inter-individual genetic variation in habitats with significant fragmentation and low population sizes [54, 56–58]. Similarly, in diploid L. dendroides from Canary Islands, despite radical habitat fragmentation and small population size, some subpopulations of have enough genetic variation to compensate for the influence of drift [59]. However, in plant populations with low genetic variability, epigenetic variation can also act as an important source of potentially adaptive phenotypic variability [11–13]. Nevertheless, the extent and importance of epigenetic variation in natural populations of sexual and agamospermous Limonium species is still largely unexplored [9, 10, 60].

Data generated in the current study unsurprisingly reveals that tetraploids have a higher level of methylation than diploids. In part this may be accounted for the increased genome size, with scope for intergenome heterozysity for methylation marks adding to that expected between homologous loci. There are also additional indirect processes that can contribute to an expected increased incidence in methylation across polyploid species. In newly formed polyploids, genomic duplications from transposons and duplication of regulatory genes, can result in high levels of cytosine methylation on the new portions of the genome [61–63]. Several methylation-based processes have been implicated in the competitive silencing of duplicated gene regions (e.g. [64]), leading to increased methylation relative to the diploid progenitor(s). The same processes also lead to the expectation of differential methylation patterning between the diploid progenitor and the polyploid offspring, partly because of imperfect duplication of genes in the neopolyploid, as was reported for the Waxy gene in Spartina[64], but also because of various systems of asymmetric homeolog silencing (e.g. [64]). Given the causal links between methylation and gene regulation [5], there is scope for such changes to influence the phenotype of the polyploids relative to their diploid relatives. Principal coordinate and genetic distance analyses performed in the current study yielded greater separation between diploid and tetraploid taxa when using epigenetic information than when using only genetic information, suggesting that ploidy levels are better separated using epigenetic information than genetic information alone. Intriguingly, this pattern of variation was mirrored by the more pronounced morphological separation between the diploid and tetraploid taxa than that seen between species at either ploidy level. Should this observation apply more broadly to other groups, it implies that epigenetic profiling may provide a useful additional tool to infer ploidy level of preserved specimens.

We found similar trends when making comparisons at the species level. Moreover, the diploid species within L. ovalifolium complex have imperfect but reasonable morphological differentiation but genetic co-variation with species identity was relatively modest. Interspecies separation was more strongly enhanced when analysis was focused on the epigenetic variation encompassed in the HpaII profiles, implying that epigenetic patterning (and associated gene silencing) may play a significant role in species separation of the group and so may have some utility for species diagnosis. In both cases, further research would be required to convert these anonymous profiles into Sequence Tagged Site epigenetic markers for robust diagnostic purposes.

Analysis of genetic/epigenetic variability using AMOVA revealed that tetraploid species present lower levels of genetic variability than diploid species. This observation could be most plausibly explained through consideration of their reproductive biology. Several previous studies of the diploids L. ovalifolium, L. nydeggeri, and the tetraploids L. dodartii, L. multiflorum and L. vulgare have provided some insights into their primary reproductive strategies [14, 31, 65]. These and other works were based on the determination of flower dimorphisms linked to a sporophytic self-incompatibility system [14, 17, 18, 28, 33, 66, 67]. In these studies diploids L. nydeggeri and L. ovalifolium were deemed probable sexual species based on their reproductive characteristics. The same applies to the tetraploid L. vulgare[14, 65]. Conversely, tetraploids L. dodartii and L. multiflorum both belonging to the Limonium binervosum group, were considered as agamospermous [31, 42, 53]. Hence, it seems most likely that the lower level of genetic diversity in the putative agamospermous tetraploids could be best explained by their apomictic reproduction mode. In other polyploid apomictic species, such as in Ranunculus sp., genetically uniform populations have been similarly observed as a consequence of this mode of reproduction [45]. This finding contrasts with the relatively high level of epigenetic variability among the tetraploids, leading us to speculate that in apomictic polyploid Limonium species, the lack of genetic variability caused by the loss of meiotic segregation could be partially compensated by enhanced epigenetic variation. Several authors have suggested similar heritable phenotypic variation due to stable epigenetic variation in the absence of genetic variation (e.g., Viola cazorlensis; Laguncularia racemosa, and in triploid asexual dandelion lineages (reviewed in [8–10])). Viewed in this context, the results of the present study add Limonium to that list.

The present work failed to support any relationship between genetic or epigenetic distance with geographic distance between populations of each species, except for a positive correlation between genetic and geographic distances among L. nydeggeri, consistent with regional equilibrium between gene flow and drift [68]. The absence of co-correlation is not unexpected for the apomictic polyploids, and can be explained by restricted gene flow between populations, founder events produced by a limited number of individuals, absence of recombination and spread of single asexual clones within populations [45]. One plausible explanation of these results considered collectively is to propose that genetic information flows between populations but that epigenetic information is mainly induced locally by the environment. Alternatively, it might be that selection pressure on epiloci is higher than on genetic loci, or simply that epiloci are plastic, in the sense that they appear and disappear depending on the environmental cues. Conversely, Mantel test analysis of the correlation between genetic and epigenetic distances between L. nydeggeri populations, which presents a regional equilibrium at genetic level, show a strong significant positive correlation between genetic and epigenetic distances. While, genetic and epigenetic distances between L. multiflorum populations, which does not present genetic regional equilibrium, showed a weak negative correlation. Again, this might suggest that on the absence of a strong gene flow between populations, environmental conditions exert a higher pressure on the fixation of epigenetic loci that cannot be masked by genetic variability introduced by sexual reproduction.

Higher correlation was found between morphometric and epigenetic differentiation than between morphometric and genetic differentiation. We therefore, suggest that epigenetic variation might be a driver of the observed phenotypic divergence between the studied taxa through intergenome silencing. We argue that the present work helps to demonstrate the importance of considering phenotypic, genetic and epigenetic variables when seeking to explain the dynamics of complex plant groups that feature hybridization, polyploidy and variable modes of reproductive biology.