Introduction

Miniaturization of plant size is a phenomenon well known from both arctic and alpine regions. Plants growing at high latitudes or altitudes can often be assigned to one of four major growth forms that constitute large parts of arctic–alpine floras (Johnson 1969; Körner 1999). These are (1) cushion plants; (2) tussock-forming graminoids; (3) diminutive, prostrate shrubs and (4) herbaceous and often rosulate perennials. All of these are characterized by a reduction in plant and organ size in comparison to relatives from lower latitudes or altitudes. This dwarfism has been interpreted as a morphological adaptation to the severe climatic conditions plants experience in arctic and alpine environments (Billings and Mooney 1968; Bliss 1971; Johnson 1969; Körner and Larcher 1988; Körner 1999). Transplant experiments have shown that dwarfism is a genetically fixed character in both alpine and arctic species and in alpine or arctic ecotypes of more widely distributed species (e.g., Turesson 1925, 1930; Körner et al. 1989; Körner 1999; Shinohara and Murakami 2006).

We here investigate the evolution of plant size in a clade of four genera of holarctic plants with both lowland and arctic–alpine species, i.e., Petasites Mill., Endocellion Turcz. ex Herder, Homogyne Cass. and Tussilago L. (Asteraceae–Senecioneae; Pelser et al. 2007), henceforth referred to as the Petasites-clade. Fully developed leaves of Tussilago are about 20–30 cm in diameter (Hegi 1929), and most species of Petasites have basal leaves of similar or larger size (Fig. 1); they can be up to 1.5 m in diameter in P. japonicus (Siebold & Zucc.) Maxim. var. giganteus F.Schmidt ex Makino (Hind and Kay 2006). The only exceptions in Petasites are P. doerfleri Hayek, P. rubellus (J.F.Gmel.) J.Toman and P. fominii Bordz., three species restricted to high altitude habitats which have small basal leaves that are <5 cm in diameter (Toman 1972; Fig. 1). Leaves of Homogyne from the Alps and other European high mountain ranges and of Endocellion from the Asian Arctic are rather small. Their rounded to cordate or ovate leaves reach only about 1–7 cm in diameter (Toman 1972; Dingwall 1976; Tutin 1976a; Fig. 1).

Fig. 1
figure 1

Leaf size variation in the Petasites-clade. a Petasites albus (after Toman 1972). b P. fragrans (drawn from Steffen 180908/1). c P. frigidus (Boyko s.n.). d P. hybridus (Steffen 130409/1). e P. japonicus (Toman 1972). f P. kablikianus (Toman 1972). g P. paradoxus (Steffen 290708/3). h P. radiatus (Toman 1972). i P. spurius (Toman 1972). j P. tatewakianus (Salokhin s.n.). k P. tricholobus (Toman 1972). l P. doerfleri (Dörfler 569). m P. rubellus (Boyko s.n.). n Tussilago farfara (Steffen 060608/1). o Endocellion glaciale (Pospepov 00-467). p E. sibiricum (Kharkevich and Buch 814b). q Homogyne alpina (Steffen 100809/4). r H. discolor (Steffen 010578). s H. sylvestris (Steffen 010572). Scale bar 1 cm

Full size image

Members of the entirely holarctic Petasites-clade are perennial, rosette-forming herbs that propagate clonally with long rhizomes; H. sylvestris (Scop.) Cass. is the only non-clonal species in this group. Tussilago is monospecific containing only T. farfara L., a species widely distributed throughout Eurasia and introduced to North America (Hegi 1929; Tutin 1976b; Kuprianova 2000; Barkley 2006). It can be found from the lowlands to high altitude habitats and is a pioneer species, colonizing open habitats such as alluvial drifts, clay pits, moraines, avalanche deposits but also roadsides with the aid of its long, slender rhizomes (Hegi 1929; Kuprianova 2000). Homogyne is endemic to the European Alpine System (sensu Ozenda 1988) and comprises three species. Homogyne alpina (L.) Cass. is widespread across the European Alpine System and grows in a wide range of habitats including montane coniferous and deciduous forests, subalpine dwarf-shrub heaths and alpine grasslands (Hegi 1929; Tutin 1976a; Aeschimann et al. 2004; Fischer et al. 2008). Homogyne discolor Cass. and H. sylvestris are confined to the peripheral ranges of the Eastern Alps, where H. discolor occurs in subalpine and alpine habitats such as subalpine dwarf-shrub heaths, alpine grasslands and snow beds, while H. sylvestris is largely restricted to coniferous and deciduous forests of the montane and subalpine altitudinal belts (Hegi 1929; Tutin 1976a; Aeschimann et al. 2004; Fischer et al. 2008). Petasites comprises 16 species (Toman 1972; Bayer et al. 2006; Chen et al. 2011) distributed mainly throughout Eurasia, with one species (P. fragrans (Vill.) C.Presl) also found in North Africa and one (P. frigidus (L.) Fr.) occurring in both northern Eurasia and North America. Species of Petasites are commonly found in moist and often disturbed habitats such as stream banks, moist subalpine and alpine slopes and meadows, marshy tundra, peat bogs and wet forest margins (Hegi 1929; Toman 1972; Dingwall 1976; Ellenberg 1996; Cherniawsky and Bayer 1998; Kuprianova 2000; Bayer et al. 2006; Chen et al. 2011). The two species of Endocellion can be found in Arctic regions from the Ural Mountains through Siberia to the Far East, where they grow in pebbly, rather dry tundra in lowland and montane regions (Toman 1972; Dingwall 1976; Kuprianova 2000).

Members of the Petasites-clade differ not only in plant size, but also in phenology and breeding system. Homogyne is the only genus with evergreen leaves, while the remaining three genera have basal leaves that develop either during (Endocellion) or after (Petasites, Tussilago) anthesis (Nordenstam 2007). Furthermore, Petasites and Endocellion are (sub-) dioecious: in Endocellion the female capitula are radiate and lack disc florets, and the male capitula are discoid with functionally male florets. Petasites has female capitula with numerous tubular to filiform female and sterile tubular or shortly radiate florets. Its male capitula have numerous tubular, functionally male florets. The latter sometimes can also have marginal female florets which make them hermaphrodite. Homogyne and Tussilago do not show this dimorphism in reproductive structures. All capitula of Homogyne have marginal female florets which are tubular and/or very shortly radiate, and hermaphrodite disc florets. Disc florets of Tussilago are tubular and functionally male, while the numerous ray florets are female. Capitula of Tussilago, Homogyne and Endocellion are either solitary or more rarely in twos or threes on rather long peduncles, whereas capitula of Petasites are numerous and form paniculate-racemose synflorescences (Nordenstam 2007). The flowering shoots arise from rhizomes of the previous year in Tussilago and Petasites and only differ in the number of capitula they develop (Troll 1939).

A molecular phylogenetic analysis of tribe Senecioneae using ITS sequence data revealed that Petasites, Endocellion, Homogyne and Tussilago form one of four monophyletic groups of subtribe Tussilagininae s.s. (Pelser et al. 2007). Within Tussilagininae s.s. the phylogenetic position of the Petasites-clade is unclear as relationships among the four monophyletic groups (Petasites-clade: Eurasia and North America; Crocidium–Tetradymia-clade: North America; Aequatorium–Arnoglossum-clade: North and South America; Ligularia–Cremanthodium–Parasenecio complex: Asia) remained largely unresolved in the ITS phylogeny by Pelser et al. (2007). However, Pelser et al. (2007) suggested that the Petasites-clade is either sister to the remainder of Tussilagininae s.s. or to the exclusively New World Crocidium-Tetradymia-clade. Pelser et al. (2007) included representatives of all genera traditionally placed in Tussilagininae (Bremer 1994) except for Digitacalia Pippen, Nelsonianthius H.Rob. & Brettell, Paracalia Cuatrec., Pippenalia McVaugh, Psacaliopsis H.Rob. & Brettell, Rugelia Shuttlew. ex Chapm., Villasenoria B.L.Clark and Yermo Dorn. All these genera with either one or only few species grow in South, Central and North America (Bolivia, Peru, Guatemala, Mexico, southern and western USA; Nordenstam 2007). Although the phylogenetic positions of these genera in the subtribe are unclear, their distribution ranges make it unlikely that they are part of the predominantly northern temperate/boreal Petasites-clade as circumscribed by Pelser et al. (2007).

We here present a phylogenetic analysis of the Petasites-clade based on nrDNA and cpDNA sequences (ITS, ndhF-rpl32 and rpl32-trnL). We use this phylogeny mainly to examine the evolution of dwarfism within the group. In particular, we will examine how many times and where and when dwarfism evolved.

Materials and methods

Taxon sampling

Of the 22 species currently recognized in Petasites, Endocellion, Homogyne and Tussilago (see Introduction) all except P. versipilus Hand.-Mazz. were included in this study. Ligularia tsangchanensis (Franch.) Hand.-Mazz., Nordenstamia kingii (H.Rob. and Cuatrec.) B.Nord. and Tetradymia canescens DC. of the Senecioneae-Tussilagininae and Adenostyles alpina (L.) Bluff & Fingerh. of the Senecioneae-Senecioninae were used as outgroups. Most of the DNA sequences produced for this study were obtained from leaf tissue samples taken from herbarium specimens from B, HAST, M, O, PE, PGFA, PMR, VLA and WU. Others were obtained from silica dried leaf tissue. Herbarium vouchers are deposited at MJG. Additional sequences were obtained from GenBank. Voucher information for all samples is listed in Table 1.

Table 1 Materials, voucher information, DNA accession numbers and GenBank accession numbers
Full size table

DNA extraction, amplification and sequencing

Total genomic DNA was extracted using the Qiagen DNeasy Plant Mini Kit (Qiagen GmbH, Hilden, Germany) following the manufacturer’s protocol. For some herbarium specimens this method did not yield DNA of sufficient quality for PCR. In these cases, total genomic DNA was isolated as follows: mortared leaf material was incubated in 800 μl AP1 buffer (Qiagen), 8 μl RNase A (Qiagen), 60 μl 2-mercaptoethanol and 60 μl proteinase K for 12 h at 42 °C. After a second incubation step for 0.5 h at 65 °C, 260 μl AP2 buffer (Qiagen) was added to the lysate, followed by 5 min incubation on ice and 5 min centrifugation at 20,000g. The supernatant was transferred into a new reaction tube and mixed with 500 μl chloroform/isoamyl alcohol (24:1), incubated for 5 min at room temperature and centrifuged for 5 min at 20,000g. The aqueous phase was transferred into a new tube and 500 μl isopropanol was added. After incubation for 45 min at 4 °C this was centrifuged for 5 min at 4 °C. After decantation the DNA pellet was diluted in 0.2 ml TE buffer, followed by one washing step with 1 ml ice-cold 100 % ethanol and 50 μl 2.5 M sodium acetate, and incubation for 1 h at −20 °C was followed by 20 min centrifugation at 20,000g at 4 °C. After decantation a second washing step with 0.4 ml algid 70 % ethanol was conducted, followed by dissolving the dried DNA pellet in 60 μl elution buffer (Qiagen). PCR amplification of the entire ITS region was performed using primers ITS A (Blattner 1999) or ITS 18S (Muir et al. 2001) and ITS 4 (White et al. 1990) or ITS 28S (Muir et al. 2001). In some cases, it was necessary to produce two overlapping fragments using primers ITS A and ITS C (Blattner 1999), ITS D (Blattner 1999) and ITS 4. PCR amplification of ITS was performed using the procedures described in Zhang et al. (2007). The ndhF-rpl32 intergenic spacer was amplified using primers rpL32-R (Shaw et al. 2007) and ndhF (Shaw et al. 2007). Primers trnL(UAG) (Shaw et al. 2007) and rpL32-F (Shaw et al. 2007) were used for the rpl32-trnL intergenic spacer. PCR amplification for both chloroplast regions was carried out following the ITS protocol differing only in the annealing temperature which was set to 56 °C. PCR products were purified using either NucleoSpin Extract II (Macherey–Nagel, Düren, Germany) or ExoSAP-IT PCR Product Cleanup (USB Corporation, Cleveland, Ohio, USA) following the manufacturers’ protocols. Cycle sequencing was carried out using the ABI Prism Dye Terminator Cycle Sequencing ready Reaction Kit (Perkin Elmer/Applied Biosystems, Foster City, California, USA) using the primers listed above and following the manufacturer’s protocol. The purified products were analyzed on an ABI 3130XL automated sequencer by ourselves and by StarSeq GmbH (Mainz, Germany).

DNA sequence alignment and phylogenetic analyses

Forward and reverse sequences were manually edited and merged into consensus sequences using Sequencher 4.1.2 (Gene Codes Corporation, Ann Arbor, Michigan, USA), and aligned manually in MacClade v.4.1 (Maddison and Maddison 2000).

Likelihood analyses were performed for the entire ITS region and for the combined cpDNA regions (ndhF-rpl32-trnL) separately. These gene trees were inspected for supported conflict between the data sets following Pirie et al. (2008). Supported conflict (bootstrap support for different positions of taxa in both trees >70 %) was found for one sample of E. glaciale (Ledeb.) J.Toman (AST24) and for E. sibiricum (J.F.Gmel.) J.Toman (see Results). These samples were removed from the dataset before nuclear and plastid DNA data sets were combined and further analyses were performed. For Maximum Likelihood (ML) and Bayesian analyses with BEAST (BB) and MrBayes (BM), respectively, the appropriate models of DNA substitution and the best partitioning of the dataset for the inference of phylogenetic relationships were estimated using PartitionFinder v.1.1.1 (Lanfear et al. 2012). Potential partitions to be tested were ITS1, 5.8S ribosomal gene, ITS2, ndhF-rpl32 and rpl32-trnL. Best fitting substitution models were chosen under the Bayesian information criterion. For all analyses we identified three partitions that are shown in Table 2 with the best fitting substitution model for each analysis. Maximum Parsimony (MP) analysis was performed with PAUP* v.4.0b10 (Swofford 2002). In this analysis of the combined data set (ITS+ndhF-rpl32-trnL) we used 1000 random addition replicates and TBR branch swapping. Bootstrap support was also calculated in PAUP* using 1000 replicates and the same settings as in the heuristic search. For the cpDNA data set parsimony-informative gaps for the ingroup were coded by hand according to the simple indel coding approach (Simmons and Ochoterena 2000). As the ITS sequences of the ingroup contained only few parsimony-informative gaps, these were not coded.

Table 2 Partitions and substitution models for different analyses found by PartitionFinder
Full size table

Maximum Likelihood tree searches and ML bootstrap searches were performed using the online version of RAxML (Stamatakis et al. 2008; available at Cipres Science Gateway, Miller et al. 2010). The substitution models given in Table 2 were used for the different partitions of the data set, with automatic halt of bootstrapping.

Bayesian analysis was performed using MrBayes version 3.2.3 (Huelsenbeck and Ronquist 2001) at Cipres Science Gateway (Miller et al. 2010). Parsimony-informative gaps were coded for the cpDNA data set as for the MP analysis. The partitions of the dataset were analyzed with the appropriate substitution models according to Table 2. The combined data set was run twice independently for 10 million Markov Chain Monte Carlo (MCMC) generations, with sampling every 1000th generation. Convergence of model parameters was examined using Tracer v.1.5 (Rambaut and Drummond 2007); a total of 1 million generations for each run were discarded as burn-in and the maximum clade credibility tree was obtained with TreeAnnotator version 1.5.4 (Rambaut and Drummond 2009) using mean heights.

Molecular clock dating

For the molecular clock dating analysis we used the ITS data set without Adenostyles alpina because Adenostyles is only distantly related to the ingroup. The dating analysis was carried out with BEAUTi/BEAST v.1.7.5 (Drummond and Rambaut 2007) using the appropriate substitution model according to Table 2. Due to the lack of internal calibration points, we took a molecular clock dating approach using published ITS substitution rates. For this, we used the minimum and maximum ITS substitution rates in Asteraceae identified by Kay et al. (2006). A uniform distribution (parameter ucld.mean) of these rates between 0.00251 and 0.00783 substitutions per site per million years was used. Starting trees were generated randomly. The BEAST analysis was run three times independently for 25 million generations to ensure that all parameters had an effective sampling size >200 and to ensure convergence of the independent runs to the same optimum. Convergence was examined using Tracer v.1.5.

Character evolution

To investigate the evolution of dwarfism, we collected data on plant size from the literature: Hegi (1929), Polunin (1959), Toman (1972), Dingwall (1976), Tutin (1976a, b), Cherniawsky and Bayer (1998), Kuprianova (2000), Bayer et al. (2006), Chen et al. (2011), and Liu and Illarionova (2011). This was supplemented by information collected from material in the following herbaria: B, K, M, MJG, O, US and W. Each taxon was coded for leaf size and number of capitula per synflorescence. Characters were coded as follows: leaves <7 cm in diameter, leaves more than 10 cm in diameter, leaves 2–50 cm in diameter; capitula single (rarely 2–3), capitula numerous. Although leaf size and number of capitula are continuous characters, we coded them as discrete characters because the material studied can be clearly assigned to the categories defined. The only exception with respect to leaf size is P. frigidus with leaf size ranging from 2 to 50 cm which we coded as a third character state. Petasites doerfleri, P. fominii, P. frigidus and P. rubellus were coded to have numerous capitula. However, these species usually only have 3–10 flowering heads, while other species of Petasites have at least 10, but usually about 40 capitula (up to 130). Although coded as having numerous capitula, the above four species with a reduced number of capitula were highlighted in our character state reconstruction. To reconstruct the evolution of leaf size and number of capitula under Maximum Parsimony (MP) and Maximum Likelihood (ML) criteria, respectively, Mesquite v.2.75 (MP; Maddison and Maddison 2011) and the StochChar module (ML; Maddison and Maddison 2006) implemented in Mesquite v.2.75 were used. Character states were reconstructed using a cladogram based on the combined data set obtained from the BM analysis, with all unsupported nodes collapsed. Phylogenetic uncertainty was considered by reconstructing the evolution of the characters on all 18,000 trees obtained by the BM analysis and mapping the results on the cladogram.

Results

Phylogenetic relationships

The ITS data set included the Petasites-clade and Ligularia tsangchanensis, Nordenstamia kingii and Tetradymia canescens of the Tussilagininae as well as Adenostyles alpina of the Senecioninae as outgroups. In total, the ITS matrix contained 62 accessions and was 648 bp long, of which 93 were variable but uninformative and 174 were parsimony-informative (alignments see Online Resource 1).

The ndhF-rpl32-trnL data matrix consisted of 48 accessions including Adenostyles alpina as outgroup. It was 1986 bp long, with 94 variable and 68 parsimony-informative characters. Indel coding added another 9 parsimony-informative characters.

The ML phylogeny of the ITS data set is largely identical with the phylogeny of the combined data set described below. The ML phylogeny of the plastid data set was poorly resolved but showed supported conflict with the ITS data set in two cases. First, in the ITS phylogeny E. sibiricum is part of a clade with P. frigidus and P. tatewakianus Kitam. (ML bootstrap support 100 %; electronic supplementary Online Resource 2), whereas E. glaciale groups with P. radiatus (J.F.Gmel.) J.Toman and P. spurius (Retz.) Rchb.f. (ML bootstrap support 100 %; Online Resource 2). In the plastid data set (Online Resource 3), one sample of E. sibiricum (AST23) groups with E. glaciale (74 %) and the other (AST71) as first branch of an unsupported clade including E. glaciale, E. sibiricum (AST23), P. spurius and P. tatewakianus. Second, E. glaciale (AST24) is part of a monophyletic E. glaciale in the plastid data set (85 %) but supported sister to P. spurius and P. radiatus in the ITS data set (95 %) which makes E. glaciale paraphyletic in relation to these two species. For further analyses, the samples of E. sibiricum and E. glaciale AST24 were removed from the data set, but this conflict between data sets will be discussed below.

The MP, ML and BM analyses of the combined data resulted in trees of similar topology with respect to well-supported branches; the BM tree is shown in Fig. 2. In the strongly supported Petasites-clade (MP/ML bootstrap support 87 %/97 %/Posterior Probability 1), generic relationships were unresolved as Homogyne (100 %/100 %/1), Tussilago (100 %/100 %/1) and Petasites including Endocellion (82 %/99 %/1) formed a trichotomy in all analyses. Within Homogyne, all three species were strongly supported, and H. alpina was well supported as sister (100 %/100 %/1) to H. discolor plus H. sylvestris (94 %/95 %/1). The Petasites/Endocellion clade (82 %/99 %/1) was not well resolved, containing a basal polytomy of seven clades. The weakly supported clade I contained P. tricholobus Franch., one accession of P. japonicus, P. formosanus Kitam. and P. rubellus (−/−/0.97), and clade II contained three accessions of P. japonicus (100 %/100 %/1). Petasites kablikianus Tausch formed a clade of its own (clade III; 100 %/100 %1). Clade IV contained P. hybridus (L.) G.Gaertn. et al., P. fragrans and P. fominii and was well supported (84 %/86 %/1). Petasites albus (L.) Gaertn., P. doerfleri and P. paradoxus (Retz.) Baumg. were part of clade V (97 %/100 %/1). Clade VI was composed of a monophyletic P. frigidus (including var. palmatus (Aiton) Cronquist) and P. tatewakianus (99 %/90 %/1). Endocellion glaciale, P. spurius and P. radiatus formed clade VII (100 %/100 %/1).

Fig. 2
figure 2

Phylogenetic relationships in the Petasites-clade. Maximum clade credibility tree of the combined data set (ITS+ndhF-rpl32-trnL), all unsupported branches are collapsed. MP and ML bootstrap values (≥70 %) and Bayesian posterior probabilities (≥0.95) are given at branches. Clades discussed are marked at their nodes

Full size image

Molecular clock dating

The BEAST analysis revealed a crown group age of the Petasites-clade [clade IX in Fig. 3] of 10.08 (4.77–18.12) million years (Ma; Fig. 3; Table 3). The crown group age of Petasites including Endocellion (VIII) was 7.91 (3.91–14.26) Ma. The estimated ages of the arctic–alpine species within the Petasites-Endocellion clade were 3.37 (1.28–6.31) Ma for P. frigidus/E. sibiricum (VI), 2.25 (0.65–4.44) Ma for P. doerfleri (V), 4.12 (1.35–7.87) Ma for P. fominii (IV), 3.28 (0.97–6.46) Ma for E. glaciale (VII), and 3.64 (1.30–6.86) Ma for P. rubellus (I). The crown group age of Homogyne (X) was 5.72 (1.82–10.88) Ma, and the split between H. discolor and H. sylvestris (XI) was 3.39 (1.05–6.69) Ma.

Fig. 3
figure 3

Chronogram of the Petasites-clade obtained under a Bayesian relaxed molecular clock using published ITS nucleotide substitution rates (Kay et al. 2006). Bars at nodes represent 95 % highest posterior densities of node ages discussed. Ages for the clades discussed (I–XI) are given in Table 3

Full size image
Table 3 Estimated ages of nodes in million years found in the BEAST analysis (BB) based on published ITS nucleotide substitution rates
Full size table

Evolution of dwarfism

The results of the reconstruction of leaf size and number of capitula using ML over the combined tree are shown in Fig. 4. Both MP and ML reconstructions revealed that large leaves (>10 cm in diameter) were ancestral (ML 97 %) in Petasites including Endocellion. Maximum Parsimony and ML analyses indicated that leaves smaller than 7 cm arose five times independently, i.e., in Homogyne (ML 100 %), P. rubellus, P. fominii, P. doerfleri and Endocellion.

Fig. 4
figure 4

Cladogram of the Petasites-clade obtained from the Bayesian analysis of the combined data set collapsing unsupported branches. Pie charts show the results of the Maximum Likelihood reconstruction of leaf size and number of capitula. Left chart blue = leaves more than 10 cm in diameter, green = leaves <7 cm in diameter, red = leaves 2–50 cm in diameter, black = equivocal probability for all three categories or node absent; right chart violet = capitula single (rarely 2–3), yellow = capitula numerous (more than 3, up to 130), black = equivocal probability for both categories or node absent. Taxa with only 3–10 capitula are marked with an asterisk. The distribution of sexes is mapped for each clade ( = female,  = male,  = hermaphrodite)

Full size image

The ancestral state for number of capitula could not be reconstructed unambiguously; the ML analysis indicated numerous capitula as ancestral (ML 77 %), but the MP result was ambiguous for the Petasites-clade. Maximum Parsimony and ML analyses indicated that single capitula evolved in Tussilago (ML 100 %) and Homogyne (ML 100 %). Ancestral capitulum number is ambiguous under MP for the clade containing E. glaciale, and the ML analysis resulted in numerous capitula as ancestral (ML 32 % probability of numerous capitula and 78 % equivocal probability of numerous and single capitula). Single capitula arose in E. glaciale, and numerous capitula in the species of Petasites of this clade. Within Petasites, a reduced number of capitula (3–10) can be observed in the arctic–alpine P. doerfleri, P. fominii, P. frigidus, and P. rubellus.

Evolution of sexual systems

Petasites and Endocellion show a sexual dimorphism, whereas Homogyne and Tussilago are hermaphrodite as are all other members of Tussilagininae (Nordenstam 2007). Within the Petasites-clade, a single evolutionary transition from hermaphrodite to (sub)dioecious capitula took place in the Petasites/Endocellion clade (Fig. 4).

Discussion

Phylogeny, biogeography and classification of Petasites, Endocellion, Homogyne and Tussilago

Our phylogenetic analysis (Fig. 2) of the combined data set supports recognition of Homogyne, Tussilago and Petasites including Endocellion as monophyletic units which should be assigned generic rank. This conclusion is not contradicted by the cpDNA-based phylogeny. However, support for intergeneric relationships is lacking. While Homogyne has always been treated as a well-circumscribed and distinct genus, species of Petasites and Endocellion have a more complicated taxonomic history in terms of generic assignment. Species of these two genera have been described in Tussilago, Petasites or Nardosmia Cass., and even today generic circumscriptions are used inconsistently. For example, Flora of the USSR (Kuprianova 2000) treats most species of Petasites and Endocellion as Nardosmia, a taxon included in Petasites at subgeneric rank by Toman (1972). Toman (1972) considered Endocellion a separate genus. On the other hand, Dingwall (1976) in Flora Europaea treated E. sibiricum as P. sibiricus (J.F.Gmel.) Dingwall. Our phylogeny supports Petasites including Endocellion as a monophyletic group.

Endocellion

In the phylogeny inferred from the ITS data set (Online Resource 2) Endocellion is diphyletic. Whereas E. sibiricum forms a clade with P. frigidus and P. tatewakianus, E. glaciale groups with P. radiatus and P. spurius. In contrast to this, Endocellion is (partly) monophyletic in the phylogeny based on the cpDNA data set where one accession of E. sibiricum groups with E. glaciale. This incongruence is probably the result of hybridization between the two genera (Wendel and Doyle 1998). Hybridization between species of Petasites is well documented. Several hybrid taxa have been described (Hegi 1929; Bogle 1968; Toman 1972; Dingwall 1976; Cherniawsky and Bayer 1998; Kuprianova 2000; Bayer et al. 2006), and the wide range of chromosome numbers found in Endocellion (n = 28, 29, 30, c. 50+, 56) and Petasites (n = 10, 14, 16, 26, 28, 29, 30, 40, c. 44, 45, 60; Nordenstam 2007) probably indicates a high frequency of polyploid hybrid formation (and additional aneuploid and/or dysploid chromosome number changes) in the two groups. Petasites frigidus, the species with which E. sibiricum groups in the ITS phylogeny, and E. sibiricum occur sympatrically in parts of Arctic Siberia and the Russian Far East (Kuprianova 2000), and some of our samples of the two species were collected in the same region. As the flowering period of P. frigidus spans from May to August, and E. sibiricum flowers from June to July (Kuprianova 2000), it is conceivable that the two species could hybridize when growing in close proximity. If this hypothesis were the correct explanation for the incongruence in our data, P. frigidus ITS would have introgressed into E. sibiricum, which functioned as the maternal parent as concluded from the cpDNA data. Monophyly of Endocellion would be supported by several morphological characters. Endocellion is separated from the rest of Petasites by solitary (rarely two or three) capitula in female plants (vs. numerous capitula in Petasites), very thin rhizomes (vs. thick and sometimes tuber-like rhizomes in Petasites; Toman 1972; Kuprianova 2000), and foliage that develops during flowering (vs. after flowering in Petasites). In conclusion, we hypothesize that the two species of Endocellion represent a monophyletic group. However, Endocellion, whether monophyletic or not, is clearly nested in Petasites and should not be treated as a separate genus. When included in Petasites, E. sibiricum must be treated as Petasites gmelinii (Turcz. ex DC.) Polunin and E. glaciale as Petasites glacialis (Ledeb.) Polunin. The nestedness of Endocellion in Petasites also implies that the evolutionary transition from hermaphrodite capitula to subdioecy/dioecy took place only once in the last common ancestor of Petasites incl. Endocellion (Fig. 4).

Petasites

Based on variation in corolla tube shape of pistillate flowers, Toman (1972) divided Petasites into three subgenera, namely Petasites (comprising P. albus, P. hybridus, P. japonicus, P. kablikianus, P. paradoxus and P. tatewakianus), Capillopetalum J.Toman (P. tricholobus, P. versipilus) and Nardosmia (Cass.) Petermann (P. doerfleri, P. fominii, P. fragrans, P. frigidus, P. radiatus, P. rubellus and P. spurius). However, this subgeneric classification is not supported by our phylogeny as these subgenera are not monophyletic. Instead, species of all three subgenera are found in different supported clades. For example, species of subg. Petasites can be found in clades I, III, IV, V and VI. We could not identify any morphological characters in support of these clades, and they do not differ consistently in geographical distribution.

Homogyne

Homogyne is a well-defined endemic of the central European high mountains, and the genus has been considered a representative of the old ‘Arcto-tertiary stock’ of the indigenous Alpine flora (Vierhapper 1923). Due to its distribution at montane altitudes, H. sylvestris was regarded as the most basal lineage of the genus by Meusel and Jäger (1992), and Merxmüller (1952) considered H. discolor to be the closest relative of H. alpina, as they mainly differ in the presence vs. absence of a dense indumentum on the lower leaf surface. In contrast, our phylogeny clearly shows that H. discolor and H. sylvestris are sister to each other, and that H. alpina is sister to these two species.

Tussilago

Our phylogeny supports Tussilago in its current circumscription as a monospecific genus. Tussilago farfara is widely distributed throughout Eurasia (Hegi 1929; Tutin 1976b; Kuprianova 2000). Interestingly, we found very little sequence variation within T. farfara although we included accessions from Europe (e.g., T. farfara AST58) and Asia (e.g., T. farfara AST41) in the analysis. Tussilago farfara is a typical pioneer species from open and ruderal habitats. It can form extensive clonal populations by vegetative propagation through rhizomes in a short period of time after wind dispersal of its achenes (Pfeiffer et al. 2008). Frequent long-distance dispersal and colonization of new habitats together with clonal reproduction might explain low genetic variation in this species.

Evolution of dwarfism

Conforming to the tendency for reduction in plant size with decreasing average temperature, as observed across altitudinal and latitudinal transects (Billings and Mooney 1968; Johnson 1969; Bliss 1971; Körner and Larcher 1988; Körner 1999), most species of the Petasites-clade occurring in high mountain areas and in the Arctic are small compared to those from the lowlands or from temperate climates. The reconstruction of ancestral character states for leaf size and capitulum number as measures of dwarfism revealed that decrease in both characters evolved several times in parallel in this lineage. Parallel evolution of dwarfism has also been reported in arctic Artemisia L. (Tkach et al. 2008) and in alpine Lysimachia L. sect. Nummularia (Hill) Klatt (Kokubugata et al. 2010). In the Petasites-clade, those species living in alpine or arctic environments have both much smaller leaves than other representatives of the group and a reduced number of capitula. This is not only true for the arctic Endocellion and the alpine Homogyne, but also for the (sub-)alpine P. doerfleri, P. fominii and P. rubellus. These three species also have leaves that are smaller than 7 cm in diameter and show reduction in capitulum number. They only have three to ten capitula, whereas other Petasites species have more than ten (up to 130 in P. hybridus) capitula (Dingwall 1976; Cherniawsky and Bayer 1998; Kuprianova 2000; Chen et al. 2011). The only exception to the general rule that all species of the Petasites-clade with both small leaves and a reduced number of capitula grow in arctic or alpine environments is H. sylvestris, which grows at montane to subalpine altitudes but has small leaves. As leaf size in Homogyne is likely to have been ancestrally small, the small leaves of H. sylvestris must be interpreted as a retention of an ancestral character. However, the leaves of H. sylvestris, although categorized as small, are larger (3–6 cm) than those of H. alpina and H. discolor (1–3 cm), and the species often has more than one capitulum (i.e., 2–3) per synflorescence. In conclusion, evolution of leaf size and capitulum number are not completely linked, as is evident in Tussilago which has large leaves but only one capitulum per stem.

The pattern of plant size reduction with increasing latitude can also be observed at the intraspecific level. Petasites frigidus is widespread throughout northern Europe, northern Asia and North America, where it is found in a wide range of habitats such as wet forests, marshy tundra, peat bogs, alpine and subalpine slopes and disturbed sites like stream banks and roadsides (Dingwall 1976; Kuprianova 2000; Bayer et al. 2006). The species shows substantial morphological variation, and leaf diameter can vary from <2 to 50 cm, and the number of capitula ranges from four to about 40 (Cherniawsky and Bayer 1998). Within P. frigidus size variation seems to be determined by latitude as individuals from the arctic have smaller leaves and a smaller number of capitula than those from lower latitudes. It is unknown whether leaf size variation in P. frigidus is the result of either phenotypic plasticity or genetic variation.

It is noteworthy that the arctic and alpine representatives of the Petasites-clade with small leaves are of somewhat similar age (2.25–5.72 Ma) when mean ages are considered. This result may imply that dwarfism in the Petasites-clade not only evolved several times independently, but that its evolution was possibly driven by the same external trigger. This could have been the dramatic global climatic cooling in the late Pliocene that resulted in the origin of arctic and alpine habitats. However, considering that our age estimates are based on ITS substitution rates, and that our mean ages have large confidence intervals, any statement about similarities in age of the different lineages must be viewed very critically.

Although all species (except H. sylvestris) with small leaves plus a reduced number of capitula grow at high altitudes or latitudes, not all species of the Petasites-clade growing at high altitudes or latitudes have small leaves and a reduced number of capitula, implying that other factors than temperature must also influence leaf size and number of capitula. An example for this is P. paradoxus from the European Alps. Although this species grows at (sub-)alpine altitudes, it has leaves measuring up to 30 cm in diameter and numerous capitula. Petasites paradoxus grows at the same altitudes in the European Alps as H. alpina and H. discolor, which are true dwarfs with leaves <5 cm in diameter and only one, rarely two or three capitula. Whereas the two species of Homogyne grow in stable habitats with a low soil nutrient content and a dense vegetation cover, such as (sub-)alpine swards, dwarf-shrub stands and coniferous forests, P. paradoxus mostly grows in often disturbed habitats with an open vegetation cover such as river banks, alluvia and scree as well as nitrophilous and tall herb communities, all of which are characterized by a higher soil nutrient content (Landolt et al. 2010). This comparison suggests that factors such as nutrient availability, habitat disturbance and/or competition can counteract the trend for decrease of leaf size and capitulum number. However, the three species of Petasites with small leaves and a small number of capitula only partly fit this pattern. Whereas P. fominii like H. alpina and H. discolor grows in alpine meadows, P. doerfleri is restricted to wet screes and P. rubellus grows on stony slopes, along stream banks and at forest margins.

Although our study allows us to conclude that, with the exception of H. sylvestris, all species with small leaves and a small number of capitula grow at high altitudes or latitudes, the reverse conclusion is unjustified. Factors governing the evolution and geographical distribution of the species investigated may be, besides temperature, nutrient availability, habitat disturbance and competition. For nutrient availability, this finding is in accordance with the conclusions drawn by Ordoñez et al. (2009), who in a global study of relationships between leaf traits, climate and soil fertility concluded that “plants with leaf traits that allow a fast use of nutrients and growth but for shorter times, like high SLA (specific leaf area) and high LNC (leaf N concentration), were found at high nutrient supply, while the reverse occurred at low nutrient supply where conservation of nutrients is arguably more important”. Detailed ecological studies of the Petasites-clade are required to conclusively identify the factors governing the evolution of the characters discussed here.