MORPHOLOGICAL AND GENOME SIZE VARIATIONS WITHIN POPULATIONS OF EDRAIANTHUS GRAMINIFOLIUS “ JUGOSLAVICUS ” ( CAMPANULACEAE ) FROM THE CENTRAL BALKAN PENINSULA

The E. graminifolius complex is widely distributed in the continental part of the central and western Balkan Peninsula and is characterized by pronounced morphological variability. Plants grow on different geological substrates, span a wide altitudinal range and inhabit heterogeneous microclimatic conditions. The aim of this study was to compare morpho-anatomical and genome size variations among 31 populations of E. graminifolius, and to correlate morphoanatomical characteristics of plants with the geomorphologic and bioclimatic characteristics of their habitats. For these purposes, multivariate statistical analyses were performed. Results showed that most of morphological variability could be explained as the adaptive responses of plants to diverse environmental conditions that accompany life at different altitudes. Populations from SE Serbia had larger genome size in respect to other investigated populations. Genome size was bigger in sympatric populations of Edraianthus then in allopatric ones. Apart from the general morphological variability, plants from the Ovčar-Kablar Gorge are particularly morphologically specific.


INTRODUCTION
The genus Edraianthus represents one of the taxonomically and biogeographically most interesting and polymorphic genera of the Balkan flora.It has drawn the attention of some well-known botanists, which resulted in three published monographs related to the morphological characteristics and taxonomy of this genus (Wettstein, 1887;Janchen, 1910;Lakušić, 1974).Recently, the genus Edraianthus was the subject of extensive molecular phylogenetic and phylogeographic (Stefanović et al., 2008;Surina et al., 2011), cytogenetic (Medjedović, 1981;Medjedović et al., 2007;Siljak-Yakovlev et al., 2010), taxonomic (Lakušić et al., 2009;Surina et al., 2009;Surina and Lakušić, 2010), and morpho-anatomical studies (Rakić, 2010;Rakić et al., 2012).The overall results of these studies confirmed the existence of extreme complexity within this genus that mainly derives from its outstanding morphological plasticity and processes of hybridization and cryptospeciation.
Phenotypic plasticity is especially pronounced within the E. graminifolius complex, which accordingly represents the most taxonomically intriguing group of the genus Edraianthus (Rakić et al. 2012).The studies on molecular phylogeny (Stefanović et al., 2008), based on chloroplast DNA sequence analysis, recognized 17 different molecularly distinct groups within the E. graminifolius complex, which, along with the recent multivariate statistical analysis of morphological parameters (Rakić, 2010;Rakić et al., 2012), showed that several morphologically distinct groups of populations exist within the E. graminifolius complex from the central Balkans.They largely correspond to taxa proposed by R. Lakušić in 1974, such as the formally non-accepted E. caricinus, E. jugoslavicus, E. vesovicii, and E. montenegrinus.
Its populations extend over a wide range of elevations from 100 to 1950 m, and inhabit microclimatically heterogeneous habitats; plants frequently grow on limestone, but can also be found on different geological substrates, such as silicate or dolomite.Published data on cpDNA sequence analysis (Stefanović et al., 2008) showed that within these populations, named E. graminifolius "jugoslavicus", there are a few molecularly distinct groups.Among them, the most specific are those populations that grow exclusively on silicate bedrock on both acid and basic (serpen-tine) soils, such as the population from Nebeske Stolice (Kopaonik Mt.), populations that inhabit the limestone cliffs of the Ovčar-Kablar Gorge in western Serbia, and populations from subalpine grasslands on the Suva Planina Mt. and surrounding areas and the gorges of Sićevo and Jerma (SE Serbia).
Taking into account the pronounced morphological heterogeneity of E. graminifolius "jugoslavicus" on the central Balkan Peninsula, the aim of this study was: 1) to quantify morphological and anatomical variation between populations of E. graminifolius "jugoslavicus" on the basis of multivariate statistics; 2) to analyze the nuclear DNA content; 3) to describe the differentiation among populations, and 4) to correlate the morpho-anatomical characteristics of plants with the geomorphologic and bioclimatic characteristics of the habitats.

Plant material
The plant material was sampled from 31 populations of E. graminifolius "jugoslavicus" from the continental part of the central Balkan Peninsula, from altitudes between 100 and 1920 m (Fig. 1, Table 1).The geographic position of each population was recorded using a hand-held Global Positioning System (GPS Garmin eTrex Vista ® C).From each of the 31 populations, 15 plant samples were used for morphological and anatomical analysis.For genome size measurements, 3-5 individuals were analyzed from each of the 16 populations.Voucher specimens are deposited in the Herbarium of the Institute of Botany and Botanical Garden, Faculty of Biology, University of Belgrade (BEOU).

Morpho-anatomical analysis
Morphometric analyses were performed on dissected plant organs (rosette and cauline leaves, involucral bracts, stem and flowers) previously well preserved in glycerol: 96% ethanol (50:50, v/v).The anatomical structure of rosette leaves was analyzed on leaf cross sections obtained on a cryostat Leica CM 1850 at -21°C.Sections were cleared in parazone, washed in water and stained in safranin (1% w/v in 50% ethanol) and alcian blue (1% w/v, aqueous).Epidermal peels for analysis of stomata were obtained by warming the leaves in glacial acetic acid: 20% H 2 O 2 (1:1, v/v); they were then thoroughly washed in water and mounted in glycerol (Jain 1976).All photographs were taken by digital camera Leica DFC295 and light microscope Leica DMLS.Leaf, stem and flower surface structures were analyzed with a Leica MZ75 stereomicroscope and JEOL JSM-6460LV scanning electron microscope.The measured morpho-anatomical characters are listed in Tables 2 and 3

Chromosome preparation
For mitotic chromosome analysis of the population from the Ovčar-Kablar Gorge, root-tip meristems of germinated seeds were pre-treated with 0.002 M 8-hydroxyquinoleine for 4 h 30 min at 16°C, and fixed in cold 3/1 (v/v) ethanol/acetic acid for 48 h.After hydrolysis in 1N HCL for 12 min at 60°C and staining in Schiff 's reagent for 2 h, the squash was realized in a drop of acetic carmine.Chromosomes were observed under a Leica DMLS light microscope and photographed by DFC295 digital camera.following Marie and Brown (1993).Cell nuclei were isolated from young leaves of at least five individuals per population.Tomato lycopersicon esculentum Mill. cv. Roma (2C=1.99 pg, Marie and Brown, 1993) was used as internal standard.The standard and investigated species leaf tissues were simultaneously chopped with a razor blade in a Petri dish in 600 µl of cold buffer (Galbraith et al., 1983), supplemented with 1% polyvinyl-pyrrolidonyl 10.000 and 5 mM sodium metabisulfite.Nuclei suspensions were filtered through 48 µm nylon mesh, treated with RNAse (2.5 U/mL) at 4°C and stained with 50 µg/mL propidium iodide (Sigma Chemical Co.St. Louis, USA).For each sample, at least 5 000 to 10 000 nuclei were measured.The 2C DNA value was calculated using the linear relationship between the fluorescent signals from stained nuclei of the investigated species and the internal standard.

DNA amounts were determined by flow cytometry
A distribution map was produced with Manifold 5.50 software (Manifold System, CDA International Ltd.).

Statistical analysis
Descriptive statistics (minimal value, mean value, maximal value, variance, standard deviation (std.dev.) and coefficient of variation (CV)) were performed for each continuous character.A principal component analysis (PCA) was performed on the complete data set (comprising 465 individuals and 59 characters of which 35 morphological and 24 anatomical characters) to show the overall morphological and anatomical variation and relationships between individuals from all populations.The hypothesis of morphological and anatomical separation of analyzed populations was tested by a canonical discriminant analysis (CDA) based on the complete data set.CDA analyses were performed on individuals, and the obtained results are represented graphically in respect to a priori defined groups.Furthermore, classificatory discriminant analysis was used in order to obtain the percentage of correctly classified individuals, based on morphological and anatomical characters, respectively.Overall dissimilarity between populations was described by Mahalanobis distances, and for the clustering, the UPGMA method was used.Qualitative characters of all investigated populations were statistically analyzed by Multiple Correspondence Analysis (MCA) and presented on a scatterplot diagram and dendrogram constructed on the basis of the UPGMA cluster method.The regression analysis (linear regression) was performed to estimate the correlations between the variation of morpho-anatomical characters of E. graminifolius "jugoslavicus" and the basic bioclimatic habitat characteristics of each population, as well as between the variation genome size and the morpho-anatomical characters.The extraction of 19 bioclimatic parameters was done with DIVA-GIS 7.5 software (Hijmans et al., 2012).Prior to the regression analysis, morpho-anatomical characters were logarithmically transformed and habitat characteristics were tested for the multicollinearity using the variance inflation factor (VIF).Bioclimatic predictors that showed significant correlations with other predictors were not used in regression analysis of the anatomical characters.The statistical significance of the linear relationship between morpho-anatomical and bioclimatic habitat characteristics was adjusted by the Bonferroni correction of multiple comparisons.Tukey's HSD of Homogenous Groups for Unequal N post-hoc test was used to analyze differences in the genome size of 16 populations.All statistical analyses were performed using the package Statistica 5.1 (StatSoft 1996).

Morphological characteristics
Detailed morphological investigations showed that E. graminifolius "jugoslavicus" is characterized by exceptional morphological variability, especially related to stem height, shape and dimensions of involucral bracts and flowers, as well as to the general presence, density and orientation of hairs (Fig. 2, Table 2).
The stem is erect to ascending, longer in plants from canyons than in the high-mountain ones, and usually slightly to moderately covered by hairs directed towards the stem base.Rosette and cauline leaves are long, linear, canaliculated and pointed at the top.The lamina of the rosette leaves is usually glabrous, sometimes with rare and short non-glandular trichomes.The adaxial leaf side is mostly glabrous, except in plants from canyon populations in which the leaves are sparsely hairy with flattened trichomes oriented towards the leaf base.On the other hand, plants from the Ovčar-Kablar Gorge are densely covered with hairs that are especially pronounced on the adaxial leaf side and directed to the leaf apex, or orientated in all directions, except towards the leaf base.Cauline leaves are ovate to widely ovate in base and elongated at the top, linear and more   venta Canyon are leaves dorsiventral, and the spongy tissue strongly dominates over the palisade one (Figs.3 a-c).
Vascular tissue is well developed and represented by 1 central and 4-13 lateral ones, the number of which directly corresponds to the leaf width.Hydathodes are situated in shallow depressions along the margins of leaves, involucral bracts and calyx lobes, and they more or less protrude above the margin (Figs.3d-e).Anomocytic stomata are more numerous at the upper than at the lower leaf side, and they are rare or absent along leaf veins and leaf margin (Fig. 3f).The highest frequency of stomata is found in leaves of high mountain populations, but also in plants from the Ovčar-Kablar Gorge, whose stomata are in addition particularly small.The largest stomata, present with low frequency, are found in leaves of plants from canyons (Morača, Jerma, Sutjeska and Derventa) as well as in plants from Kopaonik Mt. (Nebeske Stolice, approximately 2 000 m a.s.l.).
Unicellular non-glandular trichomes occur on the stem surface and along the margins of leaves,  Table 4. Results of PCA for morphological, anatomical and morpho-anatomical characters within E. graminifolius "jugoslavicus" (For explanation of character acronyms see Tables 2 and 3).characterizes plants from the Ovčar-Kablar Gorge, resulting in grayish reflection of its stem, leaves, involucral bracts and calyx.The surface of the cuticle of non-glandular trichomes has a distinct micropapillose structure (Fig. 3g).

Multivariate analysis of morpho-anatomical characters
The analysis of the coefficient of variation (CV) shows that there is a certain hierarchy in variability of the analyzed morphological and anatomical traits.Involucral bracts, leaves and photosynthetic tissues exhibit higher variability, whereas other plant organs and structures, such as flowers or stomata, exhibit a more conservative character (Tables 2 and 3).The most variable characters (CV>50%) are length of the base of the central and the outer involucral bracts (B2_L2, B3_L2), thickness of the lower palisade tissue (Lpt_h) and central vascular bundle area (CB_a).A coefficient of variation lower than 20% characterizes only the dimensions of stomata, while all other investigated characters are moderately variable (CV 20-50%).
The principal component analysis (PCA) of morphological, anatomical and morpho-anatomical characters together indicate that the structural variability of the sample is complex, since the first three derived PCA components accounted for 50.6%, 50.1% and 37.6% of the total variability, respectively (Table 4).Most of the variation of morphological parameters within the first component is due to the maximal width of the cauline leaf (Le_W1), total length of the inner involucral bract (B1_L0), maximal width of the central and outer involucral bracts (B2_W1, B3_W1), length of the calyx lobe (Ca_L1), corolla height (Co_H1) and width (Co_W1), and width of the corolla lobe (Co_W2).Among anatomical characters, most of the variation within the first component is due to the thickness of the upper and lower palisade tissues (Upt_h, Lpt_h), thickness (L_w), area (S_a), perimeter (S_c), and width of the cross section (S_w), as well as to dimensions of the stomata in the upper epidermis -perimeter (StC_ Uep), length (StL_Uep) and width (StW_Uep).The PCA of both morphological and anatomical characters together showed that characters with the strongest influence on sample variability described on the first three axes are maximal width of the cauline leaf (Le_W1), maximal width of the central and outer involucral bracts (B2_W1, B3_W1), length of the calyx lobe (Ca_L1), all corolla dimensions, as well as cross section area (S_a), perimeter (S_c) and width (S_w) (Table 4).
The canonical discriminant analysis (CDA) of morphological parameters showed that, within a relatively homogenous structure of analyzed samples, there is a fine differentiation of populations according to both their geographical distribution and distribution along the altitudinal range (Fig. 4a).The correct classification of individuals into their a priori defined groups, based on the morphological parameters, was 77.87%.Mountainous populations from central (Kopaonik Mt. -Nebeske Stolice, Bele Stene; Jabuka Mt.) and south-eastern Serbia (Suva Planina Mt.-Trem, Devojački Grob, Sokolov Kamen, Sićevo Gorge) are distributed on the negative part of the first discriminant axis (DA).Mountainous and canyon populations of western Serbia are grouped together with canyon populations from Montenegro at the positive part of the first DA, but they are separated on the second discriminant axis.The cluster analysis recognizes two main groups (Fig. 4b) and two isolated populations -from the Jerma Gorge and the Varoška River Gorge.Group I includes mountainous populations from central and SE Serbia, while group II includes mountainous and canyon populations of W Serbia and Montenegro.
The analyzed populations show a very homogenous structure at the anatomical level, without indication of their differentiation according to geographical distribution or distribution along the vertical profile (Fig. 5a).The correct classification of individuals into their a priori defined groups, based on anatomical parameters, was 66.59%.Only plants from the Derventa Canyon are anatomicaly specific due to the dorsiventral structure of their leaves, strongly developed spongy tissue and thinner palisade tissue, in respect to plants from other investigated popula-tions, which is particularly highlighted within the results of the cluster analysis (Fig. 5b).
The multiple correspondence analysis (MCA) shows that, in respect to the analyzed qualitative characters, the plants from the majority of the studied populations are similar (Fig. 6).The most frequent differences are related to the density of hair cover, shape of involucral bracts, shape of the leaf cross section and number of lateral vascular bundles.Table 5. Summary statistics of regression (R²) for independent bioclimatic variables Abbreviations: BIO3 -Isothermality, BIO4 -Temperature seasonality, BIO6 -Min Temperature of Coldest Month, BIO7 -Temperature Annual Range, BIO8 -Mean temperature of the wettest quarter, BIO9 -Mean temperature of the driest quarter, BIO12 -Annual Precipitation, BIO14 -Precipitation of Driest Month, BIO15 -Precipitation seasonality, BIO18 -Precipitation of the warmest quarter.(For explanation of character acronyms see Tables 2 and 3).Plants that are most significantly different from other analyzed populations are: a) plants from the Ovčar-Kablar Gorge, which are characterized by a large number of involucral bracts with the short and wide apex, exceptionaly well developed and compact hair cover on the surface of all vegetative organs resulting in a glaucous color of plants stem, leaves, involucral bracts and calyx, as well as by hairs on the leaf surface oriented towards the leaf apex, or orientated in all directions except towards the leaf base; b) plants from the Derventa Canyon, which are strongly shade-adapted, their leaves are dorsiventral with 1-2 palisade cell layers, 3-7 spongy cell layers (mostly five) and moderately dense leaf hair cover, and c) plants from the Jerma Canyon, which are characterized by large linear and thick rosette leaves, wide and undulate bract apex, as well as by serrate margin of the bract base.
The results of linear regression analysis are presented in Table 5.These results show that 8 of 19 bioclimatic parameters (isothermality, temperature seasonality, minimal temperature of the coldest month, temperature annual range, mean temperature of the wettest quarter, mean temperature of the driest quarter, annual precipitation, precipitation seasonality) have demonstrated statistically significant correlation with some of the analyzed morpho-anatomical characters after Bonferroni correction.Moreover, these results have shown that the orographic characteristics and bioclimatic factors related to the temperature of the habitat represent the most significant abiotic factors correlated with the morpho-anatomical differentiation of the investigated populations.The most significant correlations were found for the annual temperature range and the temperature seasonality.The morpho-anatomical variability, however, is not significantly correlated with the bioclimatic factors referring to the precipitation and humidity.According to the regression analysis, the following factors have the lowest influence on the variation of morpho-an-  As can be inferred from the regression analysis, 21 of 63 analyzed morpho-anatomical characters displayed a statistically significant correlation with some environmental factors after the Bonferroni correction.The greatest dependency in regard to the environmental factors were shown for corolla total height (Co_H1), maximal width of the central involucral bract (B2_W1), style length (St_L), calyx lobe total length (Ca_L1) and maximal width of outer involucral bract (B3_W1).

Chromosome number and genome size analyses
Because the individuals from the Ovčar-Kablar Gorge stood out from other individuals in several morphoanatomical characters, the chromosome number was determined herewith for the first time for this population.It has 2n=32 chromosomes that are small, less than 2 µm long.The nuclear DNA amount (2C value) ranged from 2.82 (population from the Varoška River Gorge) to 3.65 pg (populations from Oblik and the Suva Planina Mt.) (Table 6).The mean value of all investigated populations was 3 pg (+0.14).Nevertheless, some populations were significantly different, especially populations from the Sićevo Gorge (13), Oblik (14),and Suva Planina Mt. (16).The statistical analyses provided some interesting facts: i) the intrapopulation variation was low, except in population 14 (Table 6), and ii) the genome size was bigger in sympatric populations at the eastern boundary of E. graminifolius "jugoslavicus" distribution range in which E. jugoslavicus grows together with E. serbicus (Table 6), than in allopatric populations.Significant correlation among the genome size and morphological characters was found only between 2C value and perimeter (r = -0.68,p = 0.000) and length (r = -0.727,p = 0.000) of adaxial stomata.

DISCUSSION
The mountainous nature of the central Balkan Peninsula and diversity of its geological substrates offer an abundance of ecologically diverse habitats, which results in morphological heterogeneity of some widely distributed taxa, such as the Balkan relic and subendemic genus Edraianthus DC (Schönswetter et al., 2005;Lakušić et al., 2009;Surina et al., 2009).Populations of E. graminifolius "jugoslavicus" successfully inhabit microclimatically and pedologi- cally heterogeneous habitats, across a wide range of altitudes, from canyons to alpine belts, in the wide area of the continental part of the central Balkan Peninsula.Therefore, the variety of ecological conditions can be considered the factor that most effectively influences the morpho-anatomical adaptive response in plants from investigated populations, resulting in exceptional morphological and anatomical variability within E. graminifolius "jugoslavicus".
Most of the observed morpho-anatomical variability within E. graminifolius "jugoslavicus" is related to the plant adaptive response to different altitudes, as can be seen in the dimensions of vegetative organs that are increasingly smaller at higher altitudes.The most significant morpho-anatomical differences noticed among the analyzed plant populations are related to the shape and dimensions of the involucral bracts, all dimensions of leaves (such as length, width, and thickness), thickness of palisade and spongy tissues, as well as density of hair cover on leaf and stem surfaces.This morpho-anatomical variability of the investigated populations is found to be influenced predominantly by the temperature in the habitat.This is consistent with the fact that at higher altitudes environmental conditions become more extreme and thereby affect plant growth and physiology, resulting in the development of different plant ecotypes.The observed morphological differentiation between E. graminifolius "jugoslavicus" populations that inhabit different altitudes has already been mentioned by R. Lakušić (1974).He described several subspecies and forms that inhabit specific altitudinal ranges, such as E. jugoslavicus subsp.jugoslavicus from the montane and lower alpine belt, and E. jugoslavicus subsp.subalpinus from the subalpine and upper alpine belt.To what extent the observed morphological differences between populations of E. graminifolius "jugoslavicus" arise from phenotypic plasticity that permitted the same genotype to be maintained in different environments (Novak et al., 1991;Dewitt and Scheiner, 2004), or from genetically fixed characters (West-Eberhard, 1989;Schlichting and Pigliucci, 1998;West-Eberhard, 2003;Sultan, 2004;Garland and Kelly, 2006), remains to be resolved.
Apart from the overall morphological variability, there are four populations (from the Derventa Canyon, Varoška River Gorge, Jerma Canyon, and Ovčar-Kablar Gorge) that are particularly morphologically and anatomically specific.(1) Plants from Derventa canyon are characterized by wide involucral bracts with strongly revoluted and reflexed apex, and pronouncedly wide but thin rosette leaves.Their leaves are pronouncedly mesomorphic, with a dorsiventral leaf structure that was detected only in this population, larger epidermal cells and the largest stomata in respect to plants from all other studied populations.
(2) Plants from Varoška River Gorge have exceptionally large vegetative and generative organs, involucral bracts with pronouncedly wide cordate base, and large linear rosette leaves.(3) Plants from the Jerma Canyon are distinguished by wide-linear and thick rosette leaves, wide and undulated apex of the involucral bracts, and pronouncedly serrate margin of the bract base, as well as by some qualitative characters of inflorescence.The geographical position at the eastern boundary of E. graminifolius "jugoslavicus" distribution range emphasizes the specificity of this population.Molecular studies (Stefanović et al. 2008) as well as our studies related to genome size showed that these three populations do not differ from other investigated populations of E. graminifolius "jugoslavicus".(4) The specific features of plants from the Ovčar-Kablar Gorge are the exceptionally dense hair cover on plant surfaces resulting in their grayish color, raised trichomes directed towards the leaf apex or orientated in all directions except towards the leaf base, involucral bracts with a short apex, the smallest stomata, and, in particular, the capsule with basal lateral porate dehiscence.The results of recent molecular analysis of cpDNA sequences also showed that plants from the Ovčar-Kablar Gorge are molecularly distinct from other populations investigated herein (Stefanović et al. 2008).Their geographical isolation (northern boundary of the distribution range of E. graminifolius "jugoslavicus"), morphological differentiation, distinct qualitative characters and significant differences at the molecular level show that plants from this population merit the rank of a new species, E. canescens, that arises in the process of allopatric speciation (Lakušić et al., 2013).
As regards the genome size, certain populations showed more or less important variations in 2C DNA values.Three populations from the eastern boundary of E. graminifolius "jugoslavicus" distribution range (13 -Sićevo Gorge, 14 -Oblik, and 16 -Suva Planina Mt.) are characterized by a larger genome size, which is significantly different from the other populations.The population from Oblik showed a more pronounced intrapopulation variation in the amount of DNA.The fact that a bigger genome size was observed in populations that live in sympatry with Edraianthus serbicus could be explained by the presence of hybridization, which was recently confirmed in sympatric populations of E. wettsteinii and E. tenuifolius (Lakušić et al., 2009).However, further investigations are necessary in order to explain the phenomenon of increasing genome size of these populations.
Edraianthus species are generally characterized by equal chromosome number of 2n=32, although their genome size varies from 2.82 to 4.16 pg (Medjedović, 1981;Siljak-Yakovlev et al., 2010).This could be explained by differences in the size of their chromosomes (Caceres et al. 1998), and indicates that the process of speciation occurs through chromosomal rearrangements, without changes in the chromosome number (Bou Dagher-Kharrat et al., 2001;Levin, 2002;Livingstone and Rieseberg, 2003;Bennetzen et al., 2005).Differences in the genome size among the investigated populations could be the consequence of different environmental conditions as well (Knight and Ackerly, 2002;Lavergne et al., 2010), but only moderate negative correlation was found between the genome size and precipitation, especially in warm and dry periods of the year (r = -0.490,p = 0.000019; r = -0.592,p = 0.000).Although many authors found positive correlations between genome size and some morphological characters of plants (Masterson, 1994;Chung et al., 1998;Beaulieu et al., 2008), in E. graminifolius "jugoslavicus" significant negative correlation was observed only between 2C value and perimeter (r = -0.68,p = 0.000) and length (r = -0.727,p = 0.000) of adaxial stomata.
It can be concluded that most of the observed morpho-anatomical variability of the widely dis-tributed E. graminifolius "jugoslavicus" can be explained as a results of its potential to adapt effectively to different micro-environmental conditions.Our results indicate that there is a strong morphological differentiation within E. graminifolius "jugoslavicus" of those populations that are at the boundaries of its distribution range.These different populations, and potentially new taxa, which probably arose in the process of cryptic allopatric speciation and ancient and recent interspecific hybridization within Edraianthus, are generally followed by extensive morphological homoplasy that represents an additional problem in the resolution of opened taxonomic questions within the genus Edraianthus.Consequently, additional research related to phenotypic plasticity and inferences obtained from fast evolving DNA sequences is necessary to resolve the relationships between E. graminifolius "jugoslavicus" populations in detail and to interpret the mechanisms of their differentiation.
. All measurements were performed in Leica Q Win (Leica Microsystems).Meristic and qualitative characters used in the present analysis are as follows: Stem.Frequen-cy of non-glandular trichomes, St_d.Trichome type, St_tp.Trichome direction, St_dir.Rosette leaf.Leaf shape, Lb_sh.Non-glandular trichomes: frequency of on the leaf upper surface, UpT_d, frequency on the leaf lower surface, LoT_d.involucral bracts.Number of bracts, No_B.Shape of external bract, Br_sh.Leaf margins type, Br_c.Non-glandular trichomes: frequency of at the upper bract surface, Brup_d, direction on the upper bract surface, Brup_dir, frequency at the lower bract surface, Brlo_d, direction on the lower bract surface, Brlo_dir.inflorescence.Number of flowers, No_F.Cross section -Shape, S_sh.Number of lateral vascular bundles, Blat_ no.Number of upper palisade tissue cell layers, Upt_NoL.Number of lower palisade tissue cell layers, Lpt_NoL.Number of spongy tissue cell layers, Spt_NoL.

Fig. 6 .
Fig. 6. Results of: (a) Multiple Correspondence Analysis (MCA) of qualitative characters plotted along the first two axes; (b) Cluster Analysis of qualitative characters.

Table 1 .
Sample provenances, habitats and herbarium vouchers

Table 2 .
Descriptive statistics (N -number of samples, min -minimum value, mean -mean value of the continuous variable, maxmaximum value, std.dev.-standard deviation, CV -coefficient of variation) for analyzed morphological characters or less narrow, glabrous to moderately hairy, in some cases with slightly and rarely toothed edges.Upper cauline leaves are similar to involucral bracts.The involucral bracts (3-) 5-9 (-19) are more or less wide, ovate to cordate in base.Bract margin can be entire, undulate or serrate.Outer involucral bract is longer than inflorescence.Their surfaces are usually glabrous, with the exception of plants from the Ovčar-Kablar Gorge in which they are densely covered with hairs directed towards the apex.The rare hydathodes are situated irregularly along the margins of leaves, involucral bracts, and calyx lobes.Flowers are sessile and organized in a terminal cluster, composed of (1-) 3-7 (-14) flowers, and closely subtended by numerous involucral bracts.Calyx lobes are subulate, often with revoluted appendix.The corolla is blue violet to purple, rarely light blue to white, glabrous or ciliate at veins and corolla lobes.Its lobes are slightly revolute and reflexed at the apex.Five anthers and style do not exceed the corolla length.Capsules open by irregular apical rupture, except in plants from the Ovčar-Kablar Gorge in which they open by basal lateral pores.Seeds are ovate, brownish and flattened.4 layers of palisade cells, usually 2, whereas at the abaxial leaf side it is formed by 1 to 3 cell layers, most often just by one.Only in shade-adapted plants from the Der-

Table 3 .
Descriptive statistics (min -minimum value, mean -mean value of the continuous variable, max -maximum value, std.dev.standarddeviation, CV -coefficient of variation) for analyzed morpho-anatomical characters involucral bracts, calyx teeth and corolla lobes.The density of hairs and their orientation differ among populations.The surface of vegetative organs is mostly glabrous, but it can also be partially and slightly covered by hairs.Plant organs are rarely completely covered by trichomes, such as in plants from the Jerma Gorge, Sutjeska Canyon, Derventa Canyon and Suva Planina Mt..The exceptionaly dense hair cover

Table 5 .
Continuedatomical characters of the investigated populations: precipitation of driest month and precipitation of the warmest quarter.