Origin and genetic differentiation of pink-flowered Sorbus hybrids in the Western Carpathians

Abstract

Background and Aims Diversity of the genus Sorbus has been affected by interspecific hybridizations. Pink-flowered hybrid species have been insufficiently studied so far. They comprise bigenomic hybrid species derived from crosses S. aria s.l. × S. chamaemespilus and trigenomic ones, where S. aucuparia was involved as well. The main objective of the present study was to reconstruct their hybrid origins as well as to assess genetic distinction among several morphologically recognized hybrid species.

Methods Samples from putative maternal species and eight pink-flowered and two white-flowered hybrid species were collected in the Western Carpathians and the Sudetes. In total, 370 specimens were analysed. Six chloroplast microsatellites were used to infer parentage, whereas nuclear amplified fragment length polymorphism (AFLP) markers were employed for the identification of clones and patterns of genetic variation. Ploidy levels were estimated by flow cytometry on a subset of 140 individuals.

Key Results Genetic data supported their hybrid origins proposed based on flower and leaf morphology, and chloroplast DNA (cpDNA) revealed recurrent origins (S. caeruleomontana, S. haljamovae), even from bidirectional hybridization events (S. zuzanae). All bigenomic and trigenomic hybrid species (except triploid S. zuzanae) were found to be tetraploid. In addition to polyploidy, low genetic variation and the presence of clones within and among populations were observed, suggesting predominantly apomictic reproduction of the hybrid species. Most of the described hybrid species appeared also genetically distinct.

Conclusions The data suggest that multiple hybridization events in the Western Carpathian Sorbus have led to the formation of separate, partially reproductively isolated genetic lineages, which may or may not be discriminated morphologically. Even bidirectional hybridization can produce individuals classified to the same taxon based on phenotype. For some hybrid taxa, hybridization pathways were proposed based on their genetic proximity to parental species and differences in genome sizes.

INTRODUCTION

Hybridization is considered to be an important mechanism of evolution of new species and a bridge for gene flow among species (Arnold, 2006). When associated with chromosome doubling (allopolyploidy), it frequently leads to changes in morphology and physiology of hybrid offspring and reproductive isolation from the parents; such newly formed genetic lineages may eventually be recognized as new hybrid species. According to Mallet (2005) at least 25 % of plant species and 10 % of animal species are involved in hybridization. Soltis and Soltis (2009) even suggested that ‘perhaps all angiosperms have likely undergone at least one round of polyploidization and hybridization has been an important force in generating angiosperm species diversity’. Additionally, hybridization is also presumed to be crucial to the evolution of apomixis (asexual reproduction through seeds), which is found almost exclusively in polyploids and highly heterozygous taxa (Savidan, 2000). The capacity for apomixis may substantially trigger speciation (Jankun, 1993; Whitton et al., 2008).

Change of ploidy and ongoing hybridization are considered major drivers of diversification also in the genus Sorbus and were studied by many authors using morphological, biochemical and genetic approaches (Liljefors, 1955; Warburg and Kárpáti, 1968; Challice and Kovanda, 1978; Aldasoro et al., 1998; Nelson-Jones et al., 2002; Robertson et al., 2004, 2010; Chester et al., 2007; Feulner et al., 2014). In Europe, this genus comprises five mostly diploid and sexual species (S. aucuparia, S. chamaemespilus, S. aria, S. domestica, S. torminalis). All of them except S. domestica participate in interspecific hybridizations, giving rise to numerous hybrid species. These hybrid taxa can be divided into two major groups: (1) white-flowered nothosubgenera Soraria and Tormaria (Májovský and Bernátová, 2001) and (2) pink-flowered nothosubgenera Chamaespilaria and Chamsoraria (Májovský and Bernátová, 2001). Trigenomic hybrid species, which combine genomes of three parental species, are rarely reported. An example is S. intermedia, reported from southern Sweden and Denmark and naturalized in the British Isles, for which S. aria s.l., S. aucuparia and S. torminalis were identified as parental species (Nelson-Jones et al., 2002).

White-flowered hybrid species from the nothosubgenera Soraria and Tormaria have been well studied. Polyploidy, multiple independent origins and facultative pseudogamous apomixis (i.e. asexuality requiring pollination to initiate seed formation) were recorded in these groups (e.g. Nelson-Jones et al., 2002; Robertson et al., 2004, 2010; Chester et al., 2007; Lepší et al., 2008; Hajrudinović et al., 2015). Diploid sexual hybrids are also reported (Meyer et al., 2014). The maternal species of these hybrids are usually S. aucuparia (for Soraria) and S. torminalis (for Tormaria), whereas the paternal one is S. aria s.l. (Nelson-Jones et al., 2002; Chester et al., 2007). The opposite direction of hybridization occurred as well, but much less frequently (Oddou-Muratorio et al., 2001a; Nelson-Jones et al., 2002). There is a variety of pathways of formation of polyploid plants (Ramsey and Schemske, 1998). For white-flowered Sorbus allopolyploids, both formation of tetraploids through a triploid bridge (Nelson-Jones et al., 2002; Robertson et al., 2004) and formation of triploids via hybridization of tetraploid and diploid parents (Robertson et al., 2004; Pellicer et al., 2012) were reported. Additionally, in all Sorbus hybrids apomixis facilitates the preservation of newly formed genetic lineages and allows them to evolve into separate taxa, even in cases when sexuality becomes partially or completely restored during their evolution (Jankun, 1993; Ludwig et al., 2013).

Much less attention has been paid to the pink-flowered Sorbus hybrid species, which are usually shrubs occurring in the dwarf-pine vegetation zone at altitudes of 1000–1500 m. In this case, the parental species were inferred exclusively on the basis of morphology (Jankun, 1993; Bernátová and Májovský, 2003), with no genetic evidence provided so far. A detailed morphological study of pink-flowered hybrid species recovered two groups: bigenomic species putatively arising from crosses of S. aria s.l. × S. chamaemespilus, and trigenomic ones with a more complex origin, probably coming from hybridizations S. aria s.l. × S. chamaemespilus × S. aucuparia (Bernátová and Májovský, 2003). For the classification of trigenomic pink-flowered hybrid species, Májovský and Bernátová (2001) proposed the nothosubgenus Chamsoraria, indicating three parental subgenera involved in their origin. Little is known about their reproduction modes and ploidy levels. Chromosome numbers have been determined only for S. margittaiana (2n = 4x = 68; Jankun and Kovanda, 1989) and S. sudetica (2n = 4x = 68; Liljefors, 1955).

Here we studied seven out of eight pink-flowered Sorbus hybrid species in the Western Carpathians (Table 1) described by Jávorka (1927) and Bernátová and Májovský (2003), as well as S. sudetica from the Czech Republic. The supposed parental combinations together with names for nothosubgenera as mentioned above are presented in Fig. 1.

Table 1.

Sorbus material collected and analysed in the present study. Numbers of studied individuals are given. Localities: ST, Stratenec; PS, Poludňové Skaly; SA, Skalná Alpa; PK, Pekárová; CK, Čierny Kameň; PL, Plešovica; SL, Salatín; SN, Siná; IH, Ihrík; LJ, Labské Jámy; PpS, Pec pod Sněžkou. For details of localities (including geographical coordinates and altitudes) see Table S1

Species	Location											Total
	Slovakia									Czech Republic
	Malá Fatra		Veľká Fatra				Nízke Tatry		SGK	Krkonoše
	ST	PS	SA	PK	CK	PL	SL	SN	IH	LJ	PpS
S. aucuparia	10	4	8	3	3	0	6	2	0	0	0	36
S. aria	6	6	2	11	−	0	1	19	0	−	−	45
S. chamaemespilus	2	1	45	−	19	−	26	7	−	−	−	100
Subgenus Chamaespilaria (S. aria × S. chamaemespilus; pink-flowered)
S. zuzanae	0	0	20	−	4	−	6	6	−	−	−	36
S. haljamovae	−	6*	10	−	1	−	−	2	−	−	−	19
S. sudetica	−	−	−	−	−	−	−	−	−	15	15	30
Subgenus Chamsoraria (S. aria × S. chamaemespilus × S. aucuparia; pink-flowered)
S. montisalpae	0	−	12	−	12	−	−	−	−	−	−	24
S. margittaiana	20	−	−	−	−	−	−	−	−	−	−	20
S. atrimontis	−	−	2	−	1	−	−	−	−	−	−	3
S. caeruleomontana	−	−	−	−	−	−	−	22	−	−	−	22
S. salatini	−	−	−	−	−	−	3*	−	−	−	−	3
Subgenus Soraria (S. aria × S. aucuparia; white-flowered)
S. cf. austriaca	−	−	−	−	−	−	−	9	−	−	−	9
S. pekarovae	−	−	−	21	−	−	−	−	−	−	−	21
F 1 hybrids	−	−	−	−	−	1*	−	−	1*	−	−	2
Total	38	17	99	35	40	1	42	67	1	15	15	370

SGK, Spišsko-gemerský kras Mountains.

^*Herbarium specimen from herbarium of the Botanical Garden of the Comenius University, Blatnica.

−, the taxon has not been reported from the locality; 0, the taxon was reported from the locality but not recovered.

Fig. 1.

Scheme of relationships among the studied parental and hybrid taxa. ¹Ploidy levels reported by Warburg and Kárpáti (1968), Májovský and Uhríková (1990) and Hajrudinović et al. (2015).

The aims of this study were to (1) infer the origins of the described pink-flowered hybrids (test hypotheses postulated on the basis of morphological traits) based on flow cytometric ploidy estimations as well as biparentally (amplified fragment length polymorphisms) and maternally inherited (plastid microsatellites) molecular markers, (2) to propose plausible scenarios of their formation, and (3) to assess their distinction and differentiation, bearing in mind that multiple hybrid species were described from a single progenitor pair.

MATERIALS AND METHODS

Plant material

Sampling for DNA analysis was carried out during the flowering seasons 2009 and 2010 and the specimens were determined based on diagnostic morphological characters (Bernátová and Májovský, 2003) either directly in the field or subsequently on the herbarium specimens. For hybrids we follow the classification of Bernátová and Májovský (2003), as it offers names for most groups of the hybrid species and these names clearly imply the origin of the hybrid subgenera. Leaf samples were collected from putative parental species (S. aria, S. aucuparia, S. chamaemespilus) and seven pink-flowered hybrid species (both bigenomic and trigenomic ones) from six locations in the Western Carpathians, Slovakia (Fig. 2; Supplementary Data Table S1) covering the full previously described morphological variability. The eighth species, S. diversicolor Bernátová & Májovský, is very rare and we had only one sample available, which was not sufficient for the present study. In the case of S. salatini, the individuals were not recovered in the field and herbarium specimens were analysed. Two populations of S. sudetica from the Krkonoše (Giant Mountains) in the Czech Republic (locus classicus) were also included in the analyses. Our focus was on pink-flowered hybrids, suggesting S. chamaemespilus as one of the progenitors. White-flowered Soraria hybrid species growing at locations close to the occurrences of S. chamaemespilus in central Slovakia were included as well, as they represent potential progenitors of the trigenomic Chamsoraria hybrids. The nothosubgenus Soraria was represented by putative F₁ hybrids, S. pekarovae Májovský & Bernátová and individuals resembling S. austriaca (here denoted as S. cf. austriaca), which co-occurs with pink-flowered hybrids at Siná, Nízke Tatry Mountains. To minimize sampling of clones, only individuals not visibly connected to each other were collected. Numbers of samples analysed for each taxon are given in Table 1. In total, we analysed 370 samples classified into 14 taxa and originating from 11 localities. Collected leaves were dried in silica gel. Total genomic DNA was isolated using the CTAB method according to Doyle and Doyle (1987), which was modified for a smaller amount of plant material.

Fig. 2.

The inset map shows the location of studied Sorbus populations in Slovakia (SK) and the Czech Republic (CZ). The distribution of the sampled localities in Slovakia is shown in detail: PS, Poludňové Skaly; ST, Stratenec; PK, Pekárová; PL, Plešovica; CK, Čierny Kameň; SA, Skalná Alpa; SL, Salatín; SN, Siná; IH, Ihrík. For more details of the localities and sampled species, see Table 1 and Table S1.

Sampling for flow cytometry was carried out during the summers of 2013 and 2016. Fresh leaf samples were collected from 140 individuals at seven localities (Table 2), preserved in plastic bags and stored at 4 °C until analysis.

Table 2.

DNA ploidy levels inferred by flow cytometry. For details of the localities, see Table S1

Taxon	Locality	Abbreviation	N	Relative genome size (mean ± s.d.)	Deduced ploidy	Relative monoploid genome size (mean ± s.d.)
S. aria	Stratenec	ST	8	0·518±0·003	2x	0·259±0·002
			1	1·058	4x	0·264±0·000
	Poludňové Skaly	PS	3	0·515±0·002	2x	0·257±0·001
			1	1·048	4x	0·262
	Siná	SN	7	0·530±0·004	2x	0·265±0·002
	Salatín	SL	1	0·517	2x	0·258
S. aucuparia	Stratenec	ST	11	0·538±0·004	2x	0·269±0·002
	Poludňové Skaly	PS	4	0·528±0·004	2x	0·264±0·002
	Skalná Alpa	SA	12	0·539±0·003	2x	0·270±0·002
	Čierny Kameň	CK	4	0·534±0·002	2x	0·267±0·001
	Siná	SN	2	0·539±0·003	2x	0·270±0·002
	Salatín	SL	8	0·533±0·003	2x	0·267±0·002
S. chamaemespilus	Stratenec	ST	2	1·079±0·011	4x	0·270±0·003
	Poludňové Skaly	PS	1	1·050	4x	0·263
	Skalná Alpa	SA	4	1·090±0·006	4x	0·272±0·002
	Čierny Kameň	CK	6	1·079±0·004	4x	0·270±0·001
	Salatín	SL	10	1·065±0·004	4x	0·266±0·001
S. cf. austriaca	Siná	SN	5	1·062±0·010	4x	0·266±0·002
S. pekarovae	Pekarová	PK	1	1·073	4x	0·268
S. zuzanae	Skalná Alpa1	SA	5	0·806±0·008	3x	0·269±0·003
			1	1·078	4x	0·269
	Čierny Kameň1	CK	1	0·800	3x	0·266
	Salatín2	SL	9	0·801±0·006	3x	0·269±0·002
S. haljamovae	Siná	SN	1	1·078	4x	0·270
	Skalná Alpa	SA	4	1·070±0·006	4x	0·268±0·002
	Čierny Kameň	CK	1	1·032	4x	0·258
S. margittaiana	Stratenec	ST	14	1·074±0·005	4x	0·268±0·001
S. montisalpae	Skalná Alpa	SA	4	1·071±0·004	4x	0·268±0·001
	Čierny Kameň	CK	2	1·060±0·004	4x	0·265±0·001
S. atrimontis	Čierny Kameň	CK	1	1·060	4x	0·265
S. caeruleomontana	Siná	SN	6	1·082±0·016	4x	0·271±0·004
Total			140

¹S. chamaemespilus haplotype; ²S. aria haplotype.

Chloroplast microsatellites

Six chloroplast microsatellite (cpSSR) loci (trnT-Lpm4, trnT-Lpm3, rps16pm2, rps16pm1, rpl16pm1, trnT-Lpm1; Chester et al., 2007) were analysed. All PCRs were performed in 6-μl reactions in a GeneAmp^® PCR System 9700 (Applied Biosystems, Foster City, CA) or an iQ5 real-time PCR detection system (Bio-Rad) thermal cycler. PCR Master mix contained ∼50 ng of DNA, 3 μL of Qiagen Multiplex PCR Kit and the following concentrations of primers: 0.4 μmtrnT-Lpm4; 0·4 μmtrnT-Lpm3; 0·05 μmrps16pm2; 0·05 μmrps16pm1; 0·05 μmrpl16pm1; 0·4 μmtrnT-Lpm1. The amplification profile consisted of an initial denaturation step at 95 °C for 15 min, followed by 32 cycles with the following profile: 30 s of denaturation at 94 °C, 90 s of annealing at 50 °C and 1 min of extension at 72 °C. Final extension was for 8 min at 72 °C. Amplification products were separated on an ABI3100 DNA sequencer (Applied Biosystems) using the GeneScan 500 ROX size standard and genotypes were determined using GeneMapper software version 3.7 (Applied Biosystems).

Amplified fragment length polymorphism

Amplified fragment length polymorphism (AFLP) analysis (Vos et al., 1995) was done according to the protocol of Applied Biosystems (Applied Biosystems, 2005) with modifications as specified here. DNA restriction and adapter ligation were performed in two steps. Restriction was done in a 10-μL reaction volume, which contained 5 U of EcoRI (Promega), 2 U of MseI (New England Biolabs, Ipswich, MA, USA), 500 ng of DNA and 2 μL of 10× Tango buffer, incubated for 3 h at 37 °C. Five microlitres of ligation mix was aliquoted into each reaction. An aliquot contained 1 U of T4 DNA ligase (New England Biolabs), 1.5 μL of T4 DNA ligase buffer containing ATP, 1 μL of MseI adapter pair (50 μm) and 1 μL of EcoRI adapter pair (5 μm). Ligation was performed for 12 h at 16 °C. Products of restriction and ligation were diluted with 100 μL of ultraclean H₂O prior to PCR amplification. Preselective PCR was done in a reaction volume of 10 μL, which contained 2 U of AmpliTaq polymerase (Applied Biosystems), 1 μL of 10× buffer II, 0·6 μL of MgCl₂ (25 μm), 0·25 μL of EcoRI + A primer (10 μm), 0·25 μL of MseI + C primer (10 μm), 0·2 μL of dNTP (10 μm) and 2 μL of diluted template. A temperature profile of the preselective PCR was as follows: initial activation of polymerase at 72 °C for 2 min, 25 cycles of denaturation for 30 s at 95 °C, annealing for 30 s at 56 °C and elongation for 2 min at 72 °C, and final elongation for 10 min at 60 °C. Products of preselective PCR were diluted with 60 μL of ultraclean H₂O prior to the selective PCR. After initial primer screening, three combinations of EcoRI and MseI primers were selected for the selective PCR: EcoRI + AGG and MseI + CTT; EcoRI + AGC and MseI + CTC; and EcoRI + ACT and MseI + CTA. Conditions for the selective PCR were the same for all three primer combinations. PCR master mix (10 μL) contained 0·4 U of BioThermStar polymerase (GeneCraft, Köln, Germany), 1 μL of 10× buffer (containing MgCl₂ at 15 mm), 0·2 μL of dNTP, 1 μL of fluorescently labelled EcoRI selective primer (1 μm), 1 μL of MseI selective primer (5 μm) and 2 μL of diluted preselective PCR template. The amplification profile consisted of an initial denaturation step of 10 min at 95 °C, 36 cycles of denaturation for 30 s at 95 °C, annealing for 1 min and extension for 1.5 min at 72 °C, and final elongation for 10 min at 72 °C. The annealing temperature in the first cycle was 65 °C and in the next 12 cycles it was decreased by 0·7 °C in each cycle. Annealing temperature in the remaining 23 cycles was 56 °C. All PCRs were performed in a GeneAmp^® PCR System 9700 (Applied Biosystems) or an iQ5 real-time PCR detection system (Bio-Rad, Hercules, CA, USA) thermal cycler. Amplification products of three selective PCRs were pooled in the equal ratio 1:1:1 and separated on an ABI3100 DNA sequencer using the GeneScan 500 ROX size standard. Scoring was done using GeneMapper software v. 3.7 (Applied Biosystems). Only markers that could be reliably scored were included and genotypes were exported as a binary matrix.

Flow cytometry

DNA ploidy levels and relative genome sizes were estimated by flow cytometric analyses of fresh leaf petioles using a Partec CyFlow space (Partec, Münster, Germany) fitted with a high-power UV LED (365 nm). Leaf petioles of the analysed sample and internal standard [Glycine max ‘Polanka’, 2C DNA = 2·50 pg (Doležel et al., 1994) or Lycopersicon esculentum ‘Stupické polní tyčkové rané’, 2C DNA = 1·96 pg (Doležel et al., 1992)] were co-chopped using a razor blade in a plastic Petri dish containing 1 mL of ice-cold Otto I buffer (0·1 m citric acid, 0·5 % Tween 20; Otto, 1990). The suspension was filtered through Partec CellTrics^® 30 µm in order to remove tissue debris and incubated for at least 5 min at room temperature. Isolated nuclei in filtered suspension were stained with 1 ml of Otto II buffer (0·4 m Na₂HPO₄.12H₂O) containing the AT-specific fluorochrome 4′,6-diamidino-2-phenylindole (DAPI; 4 µg mL^–1) and β-mercaptoethanol (2 µg mL^–1). The relative fluorescence intensity was recorded for 3000 particles. Sample/standard ratios (relative genome sizes) were calculated from the means of fluorescence histograms visualized using FloMax v2.4d software (Partec, Münster, Germany). Only histograms with coefficients of variation (CVs) <5 % for the G₀/G₁ peak of the sample were considered. The sample/standard fluorescence ratios estimated using the internal standard L. esculentum were adjusted to those using G. max by multiplying the values by a coefficient of 0·808, which was based on six repeats of ratios among the two standards measured on different days.

Data analysis

cpSSRs.

Haplotypes in the cpSSR analysis were identified using the program Haplotype Analysis (Eliades and Eliades, 2009). Maternal progenitors of hybrid taxa were determined by comparing haplotypes of the parental and hybrid species within and among the locations. In order to evaluate the reliability of the analysis, 27 samples (representing 8 % of the total sample set) were replicated. To reveal phylogenetic relationships among haplotypes, a minimum spanning network was constructed using SplitsTree 4.10 (Huson and Bryant, 2006).

AFLPs.

Identification of clones in the dataset and assigning individuals to clonal groups was done using the Genotype 1.1 software (Meirmans and Van Tienderen, 2004). The program allows the user to set the threshold for the number of differences in genotypes to consider individuals to belong to one clone. The threshold was set according to the error rate (Bonin et al., 2004) derived from 54 replicates (17 % of the total sample set).

The number of genetically different groups in the dataset was inferred using Structure 2.2 (Pritchard et al., 2000). Because pink-flowered Sorbus taxa were assumed to be of hybrid origins, the admixture model was used. The number of groups K was set to 1–11 with nine repetitions for each K and a burn-in period of 200 000, and the number of generations was 1 000 000. The number of groups was determined using the ΔK method according to Evanno et al. (2005). Groups identified in this analysis were then analysed separately to determine a possible substructure in the dataset.

Species relationships were reconstructed from Jaccard’s distance matrix by the neighbour-net method using SplitsTree 4.10 (Huson and Bryant, 2006). Clade support was estimated on a neighbour-joining tree based on Jaccard’s distance coefficients employing 1000 bootstrap replicates.

In addition, the distance matrix was subjected to principal coordinate analysis using the XLSTAT software (https://www.xlstat.com).

Genome size data.

The variation of monoploid relative genome sizes among taxa and localities within taxa was tested using analysis of variance (ANOVA; procedure GLM was used because of unbalanced design; SAS, 2009). Pairwise differences between species were tested with Duncan’s test. Taxa, ploidy levels and localities represented by a single sample were excluded.

RESULTS

cpSSR analysis

All six cpSSR loci were polymorphic, and yielded 20 alleles in total. Replicate samples did not show any differences in the length of alleles. From the combinations of the alleles, 13 composite haplotypes (denoted A–M) were identified in the dataset (Table 3). The minimum spanning network (Supplementary Data Fig. S1) allowed assignation of the haplotypes into two strongly divergent groups, one containing the haplotypes restricted to S. aucuparia (A–H) and the other the haplotypes occurring in S. aria and S. chamaemespilus (I–M). Occurrence of these haplotypes within the species (both parental and hybrid ones) and respective populations is indicated in Table 3 and Table S2. Sorbus aucuparia (N = 34) was the most variable among the parental species, harbouring seven haplotypes (A, C–H). Haplotype F was the most frequent for S. aucuparia, occurring at all locations except Siná. In S. aria (N = 37) four haplotypes were found (I, K, L, M). The haplotypes K, L and M were in general evenly represented, but at particular locations some of them predominated. Haplotype I was shared by three S. aria individuals and one S. chamaemespilus individual. Sorbus chamaemespilus (N = 68) was the least variable, with two haplotypes (I, J).

Table 3.

Chloroplast microsatellite haplotypes identified in the studied Sorbus taxa, their assortment into groups and occurrence within parental and hybrid taxa. Numbers in brackets indicate the number of individuals with a particular haplotype. More details of the occurrence of the haplotypes across the studied populations are given in Table S2

Haplotype	Alleles1	Group	Taxa		Total
	L1 L2 L3 L4 L5 L6		Parental species	Hybrid taxa
White-flowered taxa
A	110 104 268 239 162 121	AU	S. aucuparia (1)		1
B	111 104 268 239 162 121	AU		F1 hybrid2 (1)	1
C	111 104 269 239 162 122	AU	S. aucuparia (6)		6
D	111 104 270 239 162 121	AU	S. aucuparia (3)		3
E	111 104 270 239 162 122	AU	S. aucuparia (1)		1
F	112 104 268 239 162 121	AU	S. aucuparia (20)	F1 hybrid2 (1)	48
				S. pekarovae (20)
				S. cf. austriaca (7)
G	112 104 269 239 162 121	AU	S. aucuparia (1)		1
H	113 104 268 239 162 121	AU	S. aucuparia (2)		2
Pink-flowered taxa
I	143 100 270 272 164 115	CH	S. chamaemespilus (1)		4
			S. aria (3)
J	143 100 271 272 164 115	CH	S. chamaemespilus (67)	S. zuzanae (22)	143
				S. sudetica (30)
				S. montisalpae (24)
K	144 100 269 272 187 115	AR	S. aria (11)	S. zuzanae (10)	60
				S. haljamovae (13)
				S. margittaiana (20)
				S. atrimontis (3)
				S. caeruleomontana (3)
L	144 100 270 272 187 115	AR	S. aria (13)	S. caeruleomontana (18)	31
M	144 100 271 272 187 115	AR	S. aria (10)	S. zuzanae (2)	22
				S. haljamovae (6)
				S. salatini (4)

AR, Aria; AU, Aucuparia; CH, Chamaemespilus.

¹Analysed cpSSR loci (Chester et al., 2007): L1, rpl16pm1; L2, rps16pm1; L3, rps16pm2; L4, trnT-L pm1; L5, trnT-L pm3; L6, trnT-L pm4.

²Putative S. aria × S. aucuparia hybrid.

One haplotype (B), which occurred in the putative F₁ hybrid S. aria × S. aucuparia from the locality Ihrík, was not found in any of the parental species, but it was clearly related to the haplotypes typical for S. aucuparia (A–H) (Table 3). Otherwise, all white-flowered hybrid species belonging to subgenus Soraria (putative F₁ hybrid from Plešovica, S. pekarovae and S. cf. austriaca) shared haplotype F with S. aucuparia.

Intraspecific haplotype variation was observed in both pink-flowered bigenomic hybrid species (subgenus Chamaespilaria): S. haljamovae contained haplotypes K and M shared with S. aria; populations of S. zuzanae in the Nízke Tatry Mountains also contained haplotypes K and M, but the populations in the Velká Fatra Mountains shared haplotype J with S. chamaemespilus. Except S. zuzanae at Salatín, all populations of the bigenomic hybrid species were monomorphic.

None of the trigenomic pink-flowered hybrid species (subgenus Chamsoraria) shared haplotypes with S. aucuparia. The populations of S. margittaiana (N = 20), S. atrimontis (N = 3) and S. salatini (N = 4) were uniform with haplotypes K, K and M, respectively (both shared with S. aria), whereas the S. montisalpae populations (N = 24) contained haplotype J (shared with S. chamaemespilus). Within-population variability was observed in S. caeruleomontana (N = 21) at Siná, Nízke Tatry Mountains, where most individuals (18) had haplotype L; the remaining three individuals had haplotype K (both shared with S. aria). The haplotypes observed in the trigenomic hybrid species often corresponded to the haplotypes found in the co-occurring bigenomic hybrid species (Supplementary Data Table S2), which makes the latter potential maternal taxa for the trigenomic hybrid species.

AFLP analysis

Three combinations of AFLP primers generated 100 clearly scorable fragments with lengths of 63–401 bp, of which 89 % were polymorphic. Samples with unreliable amplification profiles (mostly herbarium specimens) were excluded. Ninety-six fragments were shared between parental species and hybrid taxa, three occurred only in parental species and one was found only in hybrid species. The number of fragments per sample ranged from 34 (S. aucuparia, Salatín) to 79 (S. caeruleomontana, Siná). The genotyping error rate (Bonin et al., 2004) was 0·8 %. Based on this, the threshold for identifying clones was set to 1.

Big differences were observed in the proportions of clones in the populations of the three parental species (Table 4). While both S. aria (N = 39) and S. aucuparia (N = 28) were highly diverse and each individual had a distinct AFLP genotype (with the exception of two individuals of S. aucuparia from Stratenec sharing the same AFLP genotype, but not chloroplast haplotypes and thus not considered a single genetic clone), S. chamaemespilus (N = 72) appeared as largely clonal. Sixty-six out of 72 individuals analysed were assigned to a single clonal group, whereas the remaining six individuals possessed distinct genotypes, and the most common genotype occurred repeatedly at different distant locations (Supplementary Data Table S3, Fig. S2).

Table 4.

Numbers of clonal groups (unique genotypes) across the studied populations of Sorbus taxa based on AFLP and cpSSR data. Details of the composition of clonal groups are given in Table S3

Taxon	N	NC
S. aucuparia	28	28
S. aria	39	39
S. chamaemespilus	72	7
S. zuzanae	36	7
S. haljamovae	13	7
S. sudetica	27	2
S. montisalpae	24	7
S. margittaiana	20	1
S. atrimontis	3	2
S. caeruleomontana	21	9
S. salatini	3	3
S. cf. austriaca	8	2
S. pekarovae	16	1

N, number of analysed trees; N_C, number of clonal groups.

Individuals sharing the same AFLP profile occurred in every hybrid species. The population of S. pekarovae contained individuals assigned to a single clone; the same applies to S. margittaiana (Table 4). Sorbus haljamovae, S. zuzanae and S. montisalpae consisted of seven clonal groups each. Identical genotypes (considering the threshold of one different AFLP fragment) of S. haljamovae were present at Skalná Alpa and Siná (Table S3, Fig. S2). Also, the whole population of S. zuzanae at Skalná Alpa consisted of a single clone and the same genotype occurred at the nearby locality Čierny Kameň.

The Structure analysis of the complete AFLP dataset revealed a major split into two clusters (based on ΔK statistics; Supplementary Data Fig. S3a). The first cluster comprised only S. aucuparia and the second comprised S. aria, S. chamaemespilus and their bigenomic hybrid species (subgenus Chamaespilaria) (Fig. 3A). The remaining accessions showed genetic admixture between these two major clusters. Whereas in subgenus Soraria both clusters were almost equally represented, in the trigenomic hybrid species (subgenus Chamsoraria) the proportion of the S. aucuparia cluster was approximately one-third.

Fig. 3.

Flowchart and results of the Bayesian clustering (Structure) based on AFLP data. (A) Complete data set, composed of three parental species and three subgenera of hybrid species; K = 2. (B) subset of S. aria, S. chamaemespilus and hybrid species of subgenus Chamaespilaria; K = 4. (C) subset of hybrid species of subgenus Chamsoraria; K = 3. Vertical white bars demarcate species or groups, with width of corresponding boxes proportional to the number of analysed individuals. Colouring indicates each individual’s proportional cluster assignment. S. halj., S. haljamovae; S. at., S. atrimontis; S. sal., S. salatini.

To reveal potential substructure, separate analyses were done for S. aria, S. chamaemespilus and their bigenomic hybrid species on the one hand and the trigenomic hybrid species on the other. Even though in the former case ΔK peaked at K = 3 (Supplementary Data Fig. S3b, Fig. S4a), the best interpretable results gave the subdivision for K = 4 (Fig. 3B): S. aria, S. chamaemespilus, S. sudetica and S. zuzanae together with S. haljamovae. Sorbus zuzanae (with the exception of six specimens from Salatín showing equivocal, admixed assignments) and S. haljamovae were not differentiated from each other, and neither increasing the number of groups (K) nor a separate analysis of these two hybrid species allowed their differentiation.

Among the five pink-flowered trigenomic hybrid species (subgenus Chamsoraria), only three genetically different clusters (K = 3) were identified, corresponding to S. caeruleomontana, S. margittaiana and S. montisalpae (Fig. 3C, Fig. S3e). Three S. caeruleomontana specimens (also having different cpSSR haplotypes; see above) differed from the rest of the species. Sorbus atrimontis and S. salatini were not recognized as distinct, since most of their genomes were assigned to the cluster of S. montisalpae.

Additional outcomes of the Structure analyses showing a more detailed substructure or substructure at alternative numbers of groups (K) are shown in Supplementary Data Fig. S4.

Results of the neighbour-net analysis (Fig. 4) correspond quite well to the Structure outputs. The genetic division at K=2, suggesting divergence of S. aucuparia from the other two parental species and their bigenomic hybrid species (subgenus Chamaespilaria), is also evident in the neighbour net. Sorbus sudetica forms a distinct branch, but the other two hybrid species do not: S. zuzanae is split into four branches, which generally correspond to chloroplast haplotypes and location: haplotype J (localities Čierny Kameň and Skalná Alpa, both in the Veľká Fatra Mountains), haplotype M (part of the Salatín population, Nízke Tatry Mountains), haplotype K from Siná, Nízke Tatry Mountains (clustering with S. haljamovae) and haplotype K from Salatín. Bigenomic Soraria hybrid species (S. cf. austriaca and S. pekarovae) are positioned between S. aria and S. aucuparia. Bigenomic pink-flowered Chamaespilaria hybrid species (S. sudetica, S. haljamovae and S. zuzanae) are positioned between their presumed parental species, although generally closer to S. chamaemespilus. Similarly, trigenomic hybrid species (subgenus Chamsoraria) lie in the middle between all three parental species, showing higher affinity to S. aria and S. chamaemespilus than to S. aucuparia. While S. montisalpae and S. salatini form distinct branches, differentiation among the other three hybrid species is lacking.

Fig. 4.

Neighbour-net graph of the studied Sorbus taxa based on AFLP data. Chloroplast haplotypes of S. zuzanae and S. haljamovae referred to in the text are indicated in brackets. The white circle represents one sample of S. aria and the black circle one sample of S. caeruleomontana in the cluster of S. haljamovae and S. zuzanae. Bootstrap support for major clades is shown as percentages (>60 %) based on 1000 replicates.

Scattergrams of the principal coordinate analysis were consistent with the neighbour-net graph (Supplementary Data Fig. S5). All three parental species formed separate clouds in the ordination space, and their hybrids were placed between them. The trigenomic hybrids are shifted towards S. aria and S. chamaemespilus, whereas the bigenomic hybrids are placed more or less in the centre between the respective parental species. The sole exception is S. sudetica, placed slightly eccentrically from both putative parental species. Principal coordinate analysis supported the pattern of poor differentiation between S. zuzanae and S. haljamovae, albeit the former appears here as genetically more compact.

DNA ploidy

The CVs for the G₀/G₁ sample peaks ranged from 0·90 to 3·92 % (mean = 2·12 %). Three distinct classes of sample/standard fluorescence ratios were identified with means ± s.d. 0·530 ± 0·009, 0·802 ± 0·007 and 1·068 ± 0·013. As the ratio of 0·52 was previously calibrated with a diploid chromosome count (Meyer et al., 2014) the measured sample/standard ratios were suggested to correspond to DNA ploidy levels of 2x, 3x and 4x.

As expected, both S. aria and S. aucuparia were predominantly diploid; two tetraploid trees were found in S. aria (Table 2). On the other hand, S. chamaemespilus individuals from five localities were found to be tetraploid, as were almost all bigenomic and trigenomic hybrid species. The only exception among the hybrid species is S. zuzanae, where one out of the 16 analysed individuals was tetraploid; the rest were triploids (Table 2).

The calculated monoploid relative genome sizes were quite consistent and varied from 0·255 to 0·275. Among the parental species, S. aria has a significantly smaller monoploid genome compared with both S. aucuparia and S. chamaemespilus. Nevertheless, there is also significant variation in genome size among localities of each of the parental species. On the other hand, no differences were observed among local populations of S. zuzanae with different maternal contributions (Table 5).

Table 5.

Analysis of variance of the relative monoploid DNA content among the studied Sorbus taxa and localities

Variation source	d.f.	F	P
Species	8	32·48	<0·0001
Locality (species)	17	10·32	<0·0001
Error	111
Species	N	Relative monoploid genome size		Duncan groupinga	d.f.b	Fb	Pb
		Meanc	CV (%)
S. caeruleomontana	6	0·271	1·44	A	NA
S. chamaemespilus	22	0·268	1·09	B	3	23·83	<0·0001
S. margittaiana	14	0·268	0·45	B	NA
S. aucuparia	41	0·268	0·87	B	5	10·09	<0·0001
S. zuzanae 3x	15	0·268	0·85	B	2	1·26	0·3176
S. montisalpae	6	0·267	0·61	BC	1	8·35	0·0446
S. cf. austriaca	5	0·266	0·96	C	NA
S. haljamovae	7	0·265	1·72	C	NA
S. aria 2x	21	0·261	1·33	D	3	18·84	<0·0001

^aMeans designated by identical letter are not significantly different (P > 0·05).

^bAnalyses of variance among localities within a taxon.

^cOverall mean across localities.

DISCUSSION

Genetic variability of parental and hybrid Sorbus taxa and its potential sources

Our results show great differences in genetic variation levels among the three parental species S. aria, S. aucuparia and S. chamaemespilus. High genotypic richness in S. aucuparia and S. aria is in concordance with the results of previous genetic studies (Raspé and Jacquemart, 1998; Chester et al., 2007; Robertson et al., 2010). As both species reproduce predominantly sexually, are self-incompatible (Ludwig et al., 2013) and frequently form large local populations, such an outcome is expected. Much lower genetic variability was observed in S. chamaemespilus: practically the whole of the analysed material from Slovakia seems to consist of a few genotypes, one of them being very common and spread among all studied localities. Genetic variability of S. chamaemespilus has not been studied extensively so far. The population at Skalná Alpa, Veľká Fatra Mountains, was screened by Gömöry and Krajmerová (2008) using nuclear microsatellites and was found to be genetically uniform. Reanalysing this population with AFLP revealed four genotypes, one of which was also found at other localities. In contrast to a previous study reporting a diploid chromosome number in S. chamaemespilus in Slovakia (Májovský and Uhríková, 1990), the present flow-cytometric study revealed a DNA content corresponding to the tetraploid level consistently at all sampled localities. No detailed study of the reproductive system in S. chamaemespilus has been conducted so far, but, by analogy, autotetraploids within the S. aria group are generally pseudogamous apomicts (Ludwig et al., 2013). The presence of clones within localities of S. chamaemespilus can be partly attributed to vegetative reproduction through layering of branches (Kociánová et al., 2005), but the occurrence of identical genotypes at distant locations along with autopolyploidy is a strong indication of agamospermy in this species. Localities where identical genotypes of S. chamaemespilus were found were <28 km apart, which does not represent a strong barrier for seed-dispersing birds to cross (Oddou-Muratorio et al., 2001b; P. Kaňuch, Institute of Forest Ecology, Slovakia, pers. comm.). In spite of unequal genome sizes, both AFLP profiles and chloroplast data indicate that S. aria and S. chamaemespilus are closely related species, whereas S. aucuparia is more divergent

The amount of genetic variability was very different among the putative hybrid taxa. Uniform demes were observed, as well as populations with a certain amount of genetic variability, although generally substantially lower than in parental species. The presence of identical genotypes in populations of bigenomic hybrid species from distant locations (S. sudetica, S. zuzanae, S. haljamovae) supports the hypothesis that agamospermy plays a crucial role in their reproduction. The tendency to apomixis is considered to be inherited dominantly (Hand and Koltunow, 2014). Therefore, if one of the parents (S. chamaemespilus) is an apomict (as indicated by genetic uniformity and polyploidy), hybrids are very likely to be apomicts too (Robertson et al., 2004; Ludwig et al., 2013; Hajrudinović et al., 2015).

Genetic variation found in populations of pink-flowered hybrid species includes both variability between local populations of the same taxon (e.g. S. montisalpae, S. haljamovae) and variability within local populations (e.g. S. caeruleomontana, S. zuzanae). Chloroplast DNA analysis revealed recurrent origin as one of the sources of observed genetic variability of hybrids. Different chloroplast DNA (cpDNA) haplotypes were found within a locality (stenoendemic S. caeruleomontana at Siná and S. zuzanae at Salatín), as well as among occurrences of the same species (S. haljamovae and S. zuzanae), and may indicate multiple origins of these taxa. Polytopic origin is frequent in Sorbus and has already been reported in the subgenera Soraria and Tormaria (Nelson-Jones et al., 2002; Chester et al., 2007). Indeed, recurrent formation of polyploid species is not at all exceptional (Soltis and Soltis, 1999). Hybridization can potentially occur at several localities, giving rise to independent populations. Natural selection and chromosome rearrangements then result in the formation of similar morphological forms (Rieseberg et al., 1996), which are considered either polytopic species (e.g. Ashton and Abbott, 1992; Paule et al., 2012) or, in cases of greater difference, to be separate species (Steen et al., 2000).

Hybrid origin of pink-flowered Sorbus taxa

Three parental taxa (S. aria, S. aucuparia and S. chamaemespilus) were in general well differentiated both by their AFLP profiles and by their cpSSR haplotypes. The observed chloroplast haplotypes were species-specific, with the exception of haplotype I, shared by S. aria and S. chamaemespilus. The presence of this haplotype in S. aria could be an example of plastid capture or ancestral polymorphism, as suggested by Chester et al. (2007) and Oddou-Muratorio et al. (2001a) in the case of S. aria and S. torminalis plastids.

Whereas the flow-cytometric survey proved that all hybrid taxa are polyploid, their genetic intermediacy, as indicated by the AFLP-based neighbour-net analysis, gives a strong hint at their hybrid origin. This is well supported also by the Bayesian clustering (Structure): it generally renders parental species as separate groups, while hybrid species exhibit genetic admixture grossly corresponding to the expected genomic contributions of the putative parents. It must be emphasized that Structure is not designed for the analysis of polyploid complexes using dominant markers. However, in our case it gave biologically well-interpretable results consistent with the outcomes of the distance-based approach. The hybrid taxa are positioned on relatively long terminal branches in the neighbour-net graph, and are mostly resolved as separate clusters in more detailed Structure analyses of partial datasets. Thus, the nuclear data suggest that most hybrid species represent stabilized taxa (maybe with the exception of S. zuzanae; see also below), which are not products of recent hybridization events. Studies of white-flowered hybridogenous Sorbus taxa carried out (nothosubgenera Soraria and Tormaria) hitherto showed that S. aria contributed less frequently as the maternal progenitor than S. torminalis or S. aucuparia (Oddou-Muratorio et al., 2001a; Robertson et al., 2004, 2010; Chester et al., 2007; Leinemann et al., 2013; Hajrudinović et al., 2015). This was confirmed by our analysis, in which the Soraria taxa (S. pekarovae, S. cf. austriaca and the putative F₁ hybrids) shared chloroplast haplotypes exclusively with S. aucuparia. On the other hand, haplotypes typical for both S. aria and S. chamaemespilus were found in the subgenus Chamaespilaria (S. zuzanae, S. haljamovae, S. sudetica), haplotypes of S. aria even being predominant. This proves that S. chamaemespilus can produce viable pollen like the other putatively autotetraploid apomicts (Rich, 2009; Ludwig et al., 2013) and participate in hybridization as the pollen donor. Intraspecific variation in chloroplast haplotypes may suggest origin from recurrent hybridization events of S. haljamovae and S. zuzanae and bidirectional hybridization in the latter.

Sorbus sudetica was clearly distinguished from the remaining Chamaespilaria taxa by AFLP-based approaches. On the other hand, the signal for the separation of S. haljamovae from most of the morphologically quite similar and largely sympatric S. zuzanae is uncertain. The principal coordinate analysis suggests a certain distinction between the two taxa, but they were placed in a common cluster in the Bayesian analysis. The neighbour-net graph reflects intraspecific variation in S. zuzanae, associated with a likely multiple origin of this taxon, but does not indicate distinctiveness of S. haljamovae. Genetic heterogeneity of the S. zuzanae population at Salatín may be explained by its recurrent origin. Four individuals sharing the same chloroplast haplotype (K) exhibit admixed genetic (AFLP) composition, their genomes being composed almost equally of the S. aria and S. chamaemespilus genomes. Such structure was observed in recent hybrids showing morphological intermediacy between parents as well as those morphologically undetected (Schulte et al., 2010; Paule et al., 2012) but does not prove that hybridization is still ongoing. The other two individuals from this population shared a different cpSSR haplotype (M) and also differentiated AFLP profiles placed in a separate neighbour-net branch, giving strong evidence for their independent origin. Thus, it seems that at least two hybridization events gave rise to the formation of S. zuzanae at the Salatín location. Morphological traits support distinction between S. haljamovae and S. zuzanae (Bernátová and Májovský, 2003), which is in contrast to the observed genetic patterns. Such discordance could be attributed to two genetic mechanisms. The first is selection for co-adapted gene complexes. In Helianthus (Asteraceae), even with different combinations of crossings and back-crossings synthetic lineages were similar to each other and to natural hybrids (Rieseberg et al., 1996). Interactions between parental species’ genes apparently lead to similar genomic setup of hybrid taxa even when the products of hybridization differ morphologically. The second mechanism is associated with the change of ploidy levels, which may change gene expression patterns dramatically. Epigenetic mechanisms associated with polyploidy can produce different phenotypes in spite of the similarity of genomic compositions of hybrid taxa (Hegarty et al., 2011). Nonetheless, the scenarios of the formation of the Carpathian bigenomic hybrids are quite straightforward. Sorbus zuzanae is a largely apomictic triploid, and its morphological proximity to and almost identical genome size with S. chamaemespilus make it possible to suggest that it was formed by a combination of a diploid (2x) gamete (pollen or ovule) coming from the tetraploid S. chamaemespilus and a haploid (1x) gamete coming from S. aria. It must be remembered, however, that staining was done with AT-binding DAPI, and thus the contrasts in fluorescence signal intensity not reflect not only differences among taxa in nuclear DNA content, but also potential differences in chromosome structure. For the tetraploid S. haljamovae, which is morphologically closer to S. aria and has a genome size intermediate between the parental species, two pathways are possible: either fusion of a diploid (unreduced) S. aria ovule with a diploid S. chamaemespilus pollen, or a triploid bridge mediated by S. zuzanae, where a triploid S. zuzanae ovule with S. aria haplotype is fertilized by haploid S. aria pollen. The latter scenario, however, seems less plausible, given the predominant apomixis in other Sorbus triploids (Pellicer et al., 2012) and the fact that in Skalná Alpa, with a rich population of S. haljamovae (bearing haplotype K), all S. zuzanae individuals possessed haplotype J. No hypothesis can be proposed for S. sudetica, as S. chamaemespilus has become extinct in the Krkonoše Mountains; there is thus no information about the parental ploidy. For white-flowered bigenomic hybrids, information is scarce because of limited sample sizes. Nevertheless, the absence of triploids and the intermediate position in genome size and AFLP profiles between S. aria and S. aucuparia, and the presence of haplotype F in all scored individuals, support the hypothesis that both S. austriaca and S. pekarovae originated from the fusion of an unreduced S. aucuparia ovule and unreduced S. aria pollen. The absence of S. aucuparia haplotypes in trigenomic hybrids of the subgenus Chamsoraria implies that S. aucuparia or Soraria taxa could not participate in hybridizations leading to the formation of trigenomic hybrids as ovule donors. At each locality, trigenomic Chamsoraria hybrid species are accompanied by bigenomic Chamaespilaria hybrid species with identical chloroplast haplotypes, in which a certain level of genetic variability at nuclear (AFLP) loci was observed, indicating the potential for facultative sexual reproduction. Therefore, bigenomic pink-flowered hybrids must be taken into consideration as potential maternal parents in addition to S. aria and S. chamaemespilus, which means that unambiguous determination of the maternal progenitor of trigenomic hybrid species is impossible even when the cpSSR haplotype is known.

In Chamsoraria, analyses of AFLP data employing different algorithms (Bayesian versus distance-based) rendered all hybrid species genetically distinct: neighbour-net analysis separated S. montisalpae from S. salatini and the rest, whereas with Structure it was possible to distinguish between S. margittaiana, S. caeruleomontana and S. atrimontis. This means that the combination of three particular parental genomes gave rise to several distinct hybrid taxa, which can be distinguished both morphologically and genetically, especially those occurring at a single locality only (S. margittaiana, S. caeruleomontana). In spite of this divergence, however, all trigenomic hybrids remain very similar in terms of nuclear genes (AFLP) and are displayed as a relatively homogeneous group well separated from the parental taxa and bigenomic hybrids in all analyses. This is an indication (although not a proof) of identical genomic composition despite their independent origins and even different pathways of formation evidenced by different chloroplast haplotypes. Genome sizes are quite similar for trigenomic taxa and resemble S. aucuparia and S. chamaemespilus more than S. aria. This makes it possible to suggest that the S. aria genome is underrepresented in trigenomic hybrids. Detailed hybridization pathways, however, remain uncertain in this case.

For S. montisalpae, sharing a chloroplast haplotype with S. zuzanae on the Skalná Alpa site may indicate formation via triploid bridge by fusion of an unreduced triploid S. zuzanae ovule (containing two S. chamaemespilus chromosome sets) with normal haploid pollen of S. aucuparia on this site and a subsequent spread to new localities.

Sorbus caeruleomontana, restricted to a single locality, predominantly shares a chloroplast haplotype with S. aria. Sharing of minority haplotype K with S. zuzanae might be attributed to homoplasy due to backward mutation, as the S. aria-type haplotypes differ in only one 1-bp insertion/deletion at one locus. However, the difference in chloroplast haplotype corresponds perfectly with the difference in AFLP profiles as reflected in the outcome of the Structure analysis, which supports the hypothesis of multiple independent hybridization events giving rise to S. caeruleomontana. In any case, the absence of haplotype L in S. zuzanae makes the hypothesis of triploid bridge mediated by this species in the formation of S. caeruleomontana improbable. Alternatively, the lineage of S. zuzanae (or other triploid bigenomic hybrid) containing this haplotype may have become extinct at the site. Both S. montisalpae and S. caeruleomontana exhibit relatively high genotypic richness, indicating that they may be partially sexual. Allopolyploids with an unbalanced genome may regain sexuality by chromosome rearrangements driven by homologous recombination (Gaeta and Pires, 2010). Nelson-Jones et al. (2002) proposed this mechanism as the most plausible explanation of the sexuality of allotetraploid S. intermedia. The fact that the monoploid genome size of S. caeruleomontana is significantly bigger than that of all parental species supports genome rearrangements.

For S. margittaiana and the remaining trigenomic hybrids, the absence of bigenomic hybrids at the sites of occurrence (or material thereof) and/or small sample sizes do not allow the formulation of any plausible scenarios of their formation.

The inference of evolutionary pathways from any type of molecular data or from flow-cytometric estimation of ploidy levels is of course associated with limitations, as it is based solely on genetic similarity and additivity of genome sizes. Even species-specific alleles do not give evidence of parentage unless genetic variation in parental species is screened intensively within the area of interest. Moreover, the persistence of particular genetic lineages under the harsh conditions of mountain summits is temporary in many cases and several members of the hybridization pathways may have completely disappeared in the meantime or become extinct at localities where hybridization took place. To determine the genome proportions of parental species in hybrid taxa reliably, more straightforward approaches would be necessary, associated with direct determination of chromosome identity, such as fluorescent or genomic in situ hybridization or comparisons of chromosome banding patterns after staining (Marasek et al., 2004; Van Laere et al., 2010).

It must be remembered, though, that the present study relies on established, morphology-based concepts of hybrid species of Sorbus. Nevertheless, the outcomes of our study supported the assumed hybrid origins and in addition demonstrated that recurrent interspecific hybridization events may lead to the formation of local assemblages (the use of the term ‘population’ may be inappropriate in this context), which may or may not be discriminated morphologically. Yet they represent separate genetic lineages, most likely reproductively isolated, and as such potentially deserving the status of cryptic species. Nevertheless, it is questionable whether this fact should be regarded as an incentive for a taxonomical revision of the genus. Such a situation may be a rule rather than an exception in hybrid taxa (Steen et al., 2000; Schulte et al., 2010). For practical purposes, such as nature conservation or horticulture, the present concept in the studied Sorbus taxa, recognizing nothosubgenera and particular hybrid species largely based on morphology, remains useful.

SUPPLEMENTARY DATA

Supplementary data are available online at https://academic.oup.com/aob and consist of the following. Table S1: overview of sampling locations. Table S2: distribution of haplotypes in the populations of the Sorbus taxa. Table S3: assortment of individuals into clonal groups using treshold value (maximum number of common AFLP fragments) 0 and 1. Figure S1: minimum spanning network of cp-DNA haplotypes. Figure S2: geographical distribution of clonal groups defined in Table S3 (threshold 1). Figure S3: determination of the number of groups K in the STRUCTURE analysis (Evanno et al., 2005). Figure S4: additional groups and subgroups identified by the procedure STRUCTURE (Pritchard et al., 2000). Figure S5: principal coordinate analysis based on Jaccard's genetic distances between individuals.