Taxonomy of the Solidago virgaurea Group (Asteraceae) in Poland, with Special Reference to Variability along an Altitudinal Gradient

The morphological differentiation and taxonomic treatment of lowland and high-mountain morphotypes within the Solidago virgaurea group are controversial. To clarify the taxonomic status of these taxa, we conducted a morphometric analysis of 1,746 individuals from 80 localities along an altitudinal gradient from the lowlands of northern Poland to the Carpathians and Sudetes of southern Poland. Multivariate morphometric analyses, cluster analyses and principal component analyses, were used to examine the morphological differentiation within the S. virgaurea group in Poland. Canonical discriminant analysis was applied to determine the morphological characters that best discriminate among the taxa. The stability of the high-mountain Solidago minuta morphotype was tested in an experimental field established in lowland Poland; individuals transplanted from various mountain sites were cultivated at this site, and the morphotypes remained stable in terms of their floral and vegetative characters. Multivariate analyses revealed two morphologically distinct taxa in the S. virgaurea group, which correspond to lowland S. virgaurea s. str. and high-mountain S. minuta as recognised in some European floras. The most important morphological characters for distinguishing the taxa are the number of tubular florets per capitulum, inner involucral bract width and involucre height. Vegetative and inflorescence characters appear to have less taxonomic value because they changed continuously with altitude. A key for identifying S. virgaurea and S. minuta in Poland is presented.


Introduction
to the alpine zone of the Western Carpathians and Sudetes in southern Poland. Most of the population samples (56) were collected from different altitudes of the Tatra Mts., from the foothills to the peaks. Each population sample usually consisted of 25 plants, but fewer plants were collected in some localities. Plants were collected in the field from relatively small areas (in mountains, up to ca. 15 m vertically and ca. 20 m horizontally; the areas were sometimes larger in the lowlands) and from phytosociologically uniform vegetation units. The sampled populations were morphologically homogeneous; apart from the intra-population variation of the natural populations, the plants collected at particular localities were very similar to one another. The statistical distributions of their quantitative character values deviated somewhat from normal distribution but were unimodal in every case. The population samples are listed in Appendix 1, and their distributions are given in Fig. 1.
Plants were dried and preserved as herbarium specimens for morphometric analyses. All study specimens were deposited in the KRAM (herbarium of the Institute of Botany at the Polish Academy of Sciences, Krakow).

Common Garden Experiment
To test the stability of the high-mountain morphotype and compare it with that of the lowland morphotype, we transplanted six plants per population from certain S. minuta and S. virgaurea populations to an experimental field established in the lowlands of central Poland (Wola Łagowska village at 320 m a.s.l.). The main collection from the Tatra Mts. was transplanted during the first year of the study (2005), and additional plants were transplanted in 2006. This paper presents the results of three and four years of cultivation for 62 S. minuta plants from 14 localities and 47 S. virgaurea plants from 11 localities that flowered in 2009. Because some plants did not have flowering shoots in 2009, the total number of specimens in the study is less than the number of transplanted plants (see Appendix 1 for a list of localities from which the plants were transplanted).

Morphometric Analyses
Twenty-five characters (17 continuous quantitative, seven discrete quantitative and one semi-quantitative) were measured or scored for 1,746 flowering plants from 80 populations, and eight ratios were derived from these characters ( Table 1). Characters included those traditionally used for differentiating between S. minuta and S. virgaurea, as found in determination keys and floras, and others that appeared potentially useful for distinguishing the two taxa.
Multivariate analyses were performed for 32 quantitative characters (one semiquantitative character was excluded) and for a subset of nine characters consisting exclusively of capitulum and floret characters. Additional analyses on the subset of nine characters were performed to exclude vegetative characters with greater plasticity, which might reflect the habitat effect to a higher degree (e.g., thermal conditions that change along altitudinal gradients). Three floret and involucral bract characters (LFL, LW/LL and BW/BL) were not included in the subset of capitulum and floret characters due to their high variation in the populations of both taxa, which makes their separation less clear.
For some plants it was not possible to measure or score all the examined characters; therefore, the case deletion method was applied to plants with missing data values. Thus, multivariate analyses based on individuals and 32 characters were performed for 1,380 individual samples (889 S. minuta and 491 S. virgaurea) and on nine characters for 1,578 individual samples (987 S. minuta and 591 S. virgaurea).  Differentiation of the height of involucral bracts: 1 -Bract apices form approx. one row on the involucre 2 -Bract apices form approx. two rows on the involucre 3 -Bract apices form approx. three rows on the involucre 4 -Bract apices form approx. four rows on the involucre Ratio characters InfL/PH, IBL/InfL , BW/BL, LW/LL, MLLW/MLLL, ULLW/ULLL, ULL/InfL, ULL/MLL *Characters of the capitulum and florets used in a subset of nine characters in multivariate morphometric analyses. a An inflorescence branch is defined as a structure growing from a node, having the form of a capitulum stalk when only one capitulum was present, to a well-developed branch with many capitula. b Characters concerning the capitulum and florets were obtained from one well-developed capitulum from the upper part of the inflorescence but excluding the uppermost capitulum on the shoot apex due to its larger size and often caused by two or more non-separated capitula growing together. c The middle leaf was defined as growing from the middle node between the first lower leaf and the upper stem leaf. In the case of an even number of nodes, the leaf from the lower node was taken. d The upper leaf was defined as growing from a node one below the start of the inflorescence. The inflorescence starts from a node in which a developed capitulum grows on a stalk or branch. e Character scored and discussed in the text but not used in multivariate morphometric analyses. Pearson (parametric) and Spearman (non-parametric) correlation coefficients (Zar 2010) were computed for the matrix including all plants and 32 of their characters to eliminate pairs of highly correlated characters from further analyses (Legendre and Legendre 1998).
Cluster analyses based on populations (UPGMA, an unweighted pair-group method using arithmetic averages and the Ward method, using minimisation of the increase of the sum of squares; Podani 2000) were performed to generate a hypothesis on population groupings. The populations were represented by the mean values of the measured characters. The data were standardised using a zero mean and unit standard deviation, and the Euclidean distance was used to compute the secondary matrix.
Principal component analyses (PCA) (Sneath and Sokal 1973;Podani 2000) were performed on the basis of the correlation matrices of the measured characters. The analyses were run using both populations and individual plants as objects. This method was used to reduce the multidimensionality of the original character space and to display an overall variation pattern along the first two components that extracted most of the variation.
Canonical discriminant analysis (CDA), which maximises between-group differences (Klecka 1980), was performed with individual plants and all quantitative characters to determine the extent of morphological separation between the predefined groups and to evaluate characters to distinguish groups. This method was also used to test the results from cluster analyses based on population means. Two groups resolved by cluster analyses (see Results) were defined as CDA groups. Plants from three populations (T10, K86 and K88) that were resolved differently in cluster analyses (one time among populations of S. minuta, the other time among populations of S. virgaurea, depending on the classification algorithm and variables set used) were not included in the CDA calculation of the discriminant function but were present in the classifying stage using this function. Parametric classificatory discriminant analysis was performed to estimate the percentage of plants correctly assigned to predetermined groups (taxa). Discriminant analyses require a multivariate normal distribution of characters but have been demonstrated to be highly robust against deviations from this assumption (Sneath and Sokal 1973;Klecka 1980). Means, standard deviations, minima, maxima, and the 10th and 90th percentiles were computed for all quantitative characters. The analyses were performed using Statistica ver. 7 (StatSoft Inc. 2004).

Morphometric Analyses
The results of our cluster analyses based on population averages performed for two different algorithms (UPGMA and the Ward method) and two sets of characters (32 and nine characters) indicate that all 80 populations can be divided into two clusters, namely Group 1 and Group 2, which can be identified as S. virgaurea and S. minuta, respectively (Fig. 2); however, the results differ slightly between the two algorithms. The Ward method applied to the two sets of characters gave identical results in population clustering, grouping 29 populations in Group 1 and 51 populations in Group 2 (Fig. 2b,d). Unlike the Ward method, a UPGMA based on 32 characters placed population T10 (Tatra Mts., 1,159 m a.s.l.) in Group 1 (Fig. 2a) whereas a UPGMA based on nine characters placed populations K86 and K88 (Karkonosze Mts., 1,404 m a.s.l. and 1,164 m a.s.l., respectively) in Group 1, but these populations appeared to be separate from the rest of the populations in Group 1 (Fig. 2c).
In the PCA using individuals as objects, the plants from the two taxa formed two groupings; S. virgaurea is shown on the right and S. minuta on the left of the diagrams (Fig. 4a,b). The ordination of plants based on all 32 characters did not clearly separate the plant groupings within the two taxa, leaving a zone of overlap (Fig. 4a). However, the PCA based on nine capitulum and floret characters distinctly grouped the S. minuta and S. virgaurea plants, leaving only a very narrow overlap zone (Fig. 4b). The characters with the highest correlations ( Table 2) with the first axis in the set of 32 characters were involucre height (InvH), number of shoot nodes (NSN), number of inflorescence branches having more than three capitula (NB3), number of tubular florets per capitulum (NTF), number of capitula per plant (NC) and plant height (PH). The characters with the highest correlations with the first axis in the set of nine characters were the number of tubular florets per capitulum (NTF), involucre height (InvH), tubular floret length (TFL) and involucral bract length (BL).
The Pearson (parametric) and Spearman (non-parametric) correlation coefficients did not exceed 0.95 for any pair of characters in the entire dataset; therefore, all 32 quantitative characters were included in the subsequent discriminant analyses. The pair of characters for which the Pearson correlation coefficient exceeded 0.90 was ULL−ULLL (0.922); the Spearman correlation coefficient exceeded 0.90 for ULL− ULLL (0.925), NC−NCB (0.916), NCB−NB3 (0.916) and NC−NB3 (0.901) (see Table 1 for character abbreviations).
Canonical discriminant analysis based on individual plants demonstrated a strong separation between S. minuta and S. virgaurea (Fig. 5). The characters exhibiting the highest correlations with the canonical axis were the number of tubular florets (NTF), involucral bract width (WB), involucre height (InvH) and capitulum height (CH) ( Table 2) The means, standard deviations, minima, maxima, and the 10th and 90th percentiles for quantitative characters are presented in Table 3. Figure 6 shows the variation in characters with the highest contributions to separating S. minuta and S. virgaurea, as revealed by PCA and CDA. Although the minima and maxima of all characters presented in Fig. 6 overlap considerably for the two taxa, the range of means±1 standard deviation of most do not overlap at all. The involucres of S. minuta Table 2 Results of morphometric analyses of the Solidago virgaurea group A (Fig. 3a) B  Table 1.

Common Garden Experiment
Transplanted S. minuta and S. virgaurea plants were cultivated for three and four years in an experimental field in the lowlands; the results ( Fig. 7) show that the differences in morphological characters between the two taxa are conspicuous and stable. Solidago minuta and S. virgaurea differed both in vegetative (Fig. 7a,b,c) and in capitulum and floret (Fig. 7d,e,f) characters. The phenotype of S. minuta grown under lowland conditions did not exhibit a tendency to change over the cultivation period.

Discussion
Our morphological analyses support the separation of two well-delimited and morphologically distinct taxa within the Solidago virgaurea group in Poland, which correspond well with the division of the group into lowland and high-mountain taxa in European floras. The most important characters for their division are the number of tubular florets per capitulum (NTF), involucral bract width (WB) and involucre height (InvH). European floras and keys in which the high-mountain taxon is distinguished provide discriminative characters referring to the plant's habit and some characters of its capitula, involucre and florets. All the measured vegetative characters in this study described the typical lowland and high-mountain morphotypes but had less value than the capitulum characters for the identification of the two taxa. Cluster analyses and principal component analysis based on nine capitulum characters separated the groups of taxa more clearly than the analyses based on 32 characters, including many characters related to plant habit and leaves. The importance of capitulum and floret characters for delimiting the two taxa was confirmed in canonical discriminant analysis as performed for all 32 quantitative characters studied.  Numerical analyses also revealed that morphologically intermediate forms exist between S. minuta and S. virgaurea. This was shown in the cluster analyses in which, depending on the algorithm employed (UPGMA, Ward method), populations T10 (Tatra Mts., 1,159 m a.s.l.) and K88 (Karkonosze Mts., 1,164 m a.s.l.) were classified differently. The different classification of population K86 (Karkonosze Mts., 1,404 m a.s.l.) was most likely explained by the small capitula of plants growing at the high altitudes of the Sudetes. The existence of intermediate plants was also suggested by the overlapping of plants in the PCA ordination diagrams and in the CDA histogram . However, the zone in which plants of the two taxa overlapped by morphology was relatively narrow and did not obscure the separation of two well-defined groups.
In the Flora Europaea, the author of the genus Solidago (McNeill 1976) did not clearly distinguish between high-mountain and lowland taxa. McNeill presented a morphological description and the values of some quantitative characters for S. virgaurea sensu lato but provided values for only two of them (stem height and involucre height) for subsp. minuta, which were completely within the range of variation of S. virgaurea s.l., making it impossible to identify lowland and high-mountain taxa on that basis.  One of the characters distinguishing the high-mountain from the lowland taxon of the S. virgaurea group is capitulum size. This character was noted in the first description of the high-mountain morphotype (Virga Aurea Omnium minima Floribus maximis, in Hermann 1698) and was subsequently repeated in floras and keys. The character was later quantitatively expressed as capitulum diameter, which was measured to be 10-15 mm for the lowland and 15-20 mm for the high-mountain taxon (Hess et al. 1972;Dostál 1989;Rothmaler 1994;Slavík 2004). In this work, however, the capitulum size is expressed as the number of florets, a character not biased by herbarium specimen pressing and one that is widely used in taxonomic works on the genus Solidago (e.g., Weber 1997;Nishizawa et al. 2001) and on other taxa from the Asteraceae family (e.g., Hodálová 1999;Španiel et al. 2008). Our results show that S. minuta has considerably more florets than S. virgaurea, especially tubular florets ( Table 3, key below). Another important character for expressing capitulum size as given in some European floras for discriminating between S. virgaurea and S. minuta is involucre height; for the lowland and high-mountain taxa, these were determined to be (4.5) 5-7 mm and 7-9 (10) mm, respectively (Rostański 1971;Hess et al. 1972;Wagenitz 1979;Slavík 2004). Our results approximately match these values ( Table 3). The range overlap of involucre height for the two taxa is partly due to the small size of the capitula of some S. minuta plants growing in the Tatra Mts. at elevations greater than 2,000 m a.s.l.
The results of our morphometric study differ from data presented by Slavík (2004) in the Flora of the Czech Republic. The discrepancies concern morphological characters of involucral bracts, 11 ligulate florets and 26 tubular florets for S. minuta. These differences are important because they concern diagnostic characters. Because the tubular and ligulate florets do not differ greatly in size between S. virgaurea and S. minuta, the data presented by Slavík (2004) appear to imply that the lowland taxon has a capitulum slightly larger or equal in size to that of the high-mountain taxon. This, however, would contradict not only our results but also the data presented in many European floras (mentioned above) and in the Flora of the Czech Republic (Slavík 2004). Morphometric revision of the plant material of the S. virgaurea group from the Czech Republic is required to clarify this issue.
It is worth noting that Nishizawa et al. (2001) found analogous morphological patterns for other taxa of the Solidago virgaurea complex along an elevation gradient in Japan, that is, lowland S. virgaurea subsp. asiatica Kitam. and high-mountain S. virgaurea subsp. leiocarpa (Benth.) Hultén. The high-mountain taxon exhibited a greater number of tubular florets per capitulum than the lowland taxon, and the number of tubular florets per capitulum was found to be an important diagnostic character for the lowland and high-mountain taxa.
Although not used in numerical analyses, the differentiation of involucral bract height (DHB), a semi-quantitative character of the capitulum, proved useful for identifying taxa. The values we obtained for this character in S. virgaurea and S. minuta (see Results) agree with the characteristics given by Wagenitz (1979).
We did not use the morphological characters of the basal and lowest stem leaves in this study, which had previously been used to distinguish the two taxa (Rostański 1971) because they are already wilted at the flowering stage. In our material, basal leaves were present only in 22 % of the specimens, and the lowest stem leaves were present in 41 %.
During three and four years of the S. minuta and S. virgaurea cultivation in the lowland plot, both taxa showed stable phenotypes, which must therefore have a genetic background. None of the observed morphological characters in S. minuta that were transplanted from high-mountain altitudes to the lowlands changed during the cultivation period. In cultivating transplanted S. minuta, we allowed for the possibility that high-mountain plants exhibit floral preformation; that is, buds may be formed underground one, two or even three years before starting their development (Billings 1974;Körner 2003). The very longest preformation period was reported to be four years in Polygonum viviparum (Diggle 1997). Assessing the characters of plants grown in lowland conditions from buds formed previously in an alpine environment may be a potential source of methodological error. To rule out this possibility, we extended the cultivation period to four years for some populations. During cultivation, the plants of both taxa also exhibited differences in phenology. The flowering of S. minuta lasted for approximately one month from mid-May to mid-June and that of S. virgaurea from the lowlands lasted approximately two months from mid-July to mid-September. However, S. virgaurea plants transplanted from low altitudes in the Tatra Mts. (populations T19, 942 m a.s.l. and T53, 965 m a.s.l.) flowered during a time that was intermediate between those of lowland S. virgaurea and S. minuta, starting approximately in mid-June and ending at the beginning of August.
Altitudinal vicariance exhibits a different pattern in different taxa groups. For example, Marhold (1992) found in Cardamine amara from the Carpathians and Sudetes that lowland-lower montane C. amara subsp. amara is replaced in the mountains by subalpine-upper montane C. amara subsp. opicii with an overlap zone at the altitude of 850-1,350 m a.s.l.; in this group, intermediate types besides the typical morphotypes were also frequently noted. In the Caltha palustris-C. laeta pair, Cieślak (2004) found that lowland C. palustris reached an altitude of 1,300 m a.s.l. in the Carpathians whereas mountain C. laeta descended from the subalpine and alpine belts to 300 m a.s.l. in the lowlands; therefore, the two taxa have a very wide altitudinal zone of common occurrence. Filipová and Krahulec (2006) found in the pair of Anthoxanthum odoratum-A. alpinum in the Karkonosze Mts. (Sudetes) that lowland-lower montane A. odoratum and subalpine-alpine A. alpinum overlap at 900-1,290 m a.s.l., but A. alpinum was found to descend in favourable habitats to as low as 750 m a.s.l. In the Solidago virgaurea group, lowland S. virgaurea stays in mountains mainly in the lower montane belt and occasionally reaches lower elevations of the upper montane belt (1,337 m a.s.l. in the Bieszczady Mts., 1,318 m a.s.l. in the Tatra Mts.); Solidago minuta, the range of which is above the timber line in the subalpine and alpine belts, descends to the upper montane belt and rarely to the upper part of the lower montane belt (1,073 m a.s.l. in the Tatra Mts.). Thus, the zone of altitudinal overlap for S. virgaurea and S. minuta is quite narrow. According to a classification of vascular plant altitudinal ranges in the Carpathians (Mirek 1989), S. virgaurea can be classified as lowland-lower montane, and S. minuta classified as subalpine-alpine. For all pairs of altitudinal vicariants mentioned above, the common pattern is that lowland taxa reach the lower montane belt or lower elevations of the upper montane belt. Thus, the width of the overlap zone depends mainly on the altitudinal ranges of the alpine taxa.
In this work we adopted and consistently applied a species rank for S. virgaurea and S. minuta by following their author, Linnaeus. We share the view of Porter (1893) that the divergence between lowland S. virgaurea and S. minuta (noted there as S. alpestris) "is so wide that it may be well counted a good species". In our opinion, the differences between S. virgaurea and S. minuta, with their many morphological characters, phenology, different distributions along the altitudinal gradient and related differences in occupied habitats, fully justify treating them as separate species. All of the morphological characters that we studied in the two taxa proved to be stable in cultivation. Typical Solidago virgaurea and S. minuta plants have very different habits, which are not related to the effects of species plasticity (ecological modifications) but must have a genetic background. The existence of intermediate populations between S. virgaurea and S. minuta in the contact zone of the two taxa, which are most likely the result of hybridisation, did not prevent us from distinguishing morphologically distinct taxa. This acknowledgment does not constitute a case against treating taxa at the species level, as for example, in the Senecio nemorensis group (cf. Hodálová and Marhold 1998;Hodálová 1999).

Key to the Species of the Solidago virgaurea Group Occurring in Poland
Character values given in the key represent 10th and 90th percentiles; those in parentheses represent minima and maxima.