In our carpological and molecular phylogenetic analyses, we included samples of the genus Scorzonera in its widest sense, as well as representatives of all other genera of the subtribe Scorzonerinae (Kilian et al. 2009b), to account for the previous findings that Scorzonera in all traditional circumscriptions is apparently polyphyletic. Leaf samples for DNA isolation and achene samples for carpological analysis were taken from live plants documented by herbarium specimens deposited in MW and from herbarium specimens of the herbaria B, BM, BR, E, FRU, G, HUJ, LE, MHA and MW with the permission of the curators. All samples used are listed in Table 1.

Total DNA was extracted from 30 mg of dried plant material. Samples were manually homogenised in a paper envelope and a modification of the CTAB-method of Doyle and Doyle (1987) by Krinitsyna et al. (2015) was used. After lysis with CTAB-lysis buffer (2% CTAB, 0.1 M Tris-HCl (pH 8.0), 1.4 M NaCl, 20 mM EDTA (pH 8.0), 1% PVP, 0.2% 2-mercaptoethanol, 0.1 mM DTT) for 1 hour at 60 °C, the mixture was centrifuged for 5 min at 13,400 rpm (12.080 g) and the supernatant was transferred to a new tube. An equal volume of propyl alcohol was added to the lysate and the samples were incubated at room temperature for 10 min and centrifuged again for 10 min as above. The supernatant was discarded and the precipitate was washed twice using 80% ethanol, dried for 3–5 min and dissolved in 30 µl of deionised water. The stock solution of magnetic particles (Agencourt AMPure XP, Beckman Coulter) was diluted and thoroughly mixed with the buffer (18% PEG-8000, 1M NaCl, 10mM Tris-HCl ph 8.0, 1mM EDTA pH 8.0) in the ratio of 1 to 3 (v/v). Then DNA samples were purified with the suspension of magnetic beads at a ratio of 1:1 (v/v). Concentration and purity of DNA samples were assessed by OD 260/280 and OD 260/230 ratios on the NanoPhotometer N60-Touch (Implen, Germany). DNA samples were normalised to 10 ng/µl before sample preparation. Then 5 µl (50 ng) of each normalised sample was used for PCR and library preparation.

Sequences of three markers were used for the molecular phylogenetic analyses: (1) the nrITS region (including ITS1, 5.8S rRNA gene and ITS2); (2) a ~570 bp fragment of the plastid DNA ribulose 1,5-bisphosphate carboxylase/oxygenase large subunit (rbcL) gene; (3) a 750 bp fragment of the plastid DNA maturase K gene (matK).

The nrITS region was sequenced on the MiSeq (Illumina, USA). A two-step PCR method was used for library preparation: first-stage PCR using fusion primers containing the primer sequences from Baldwin (1992) and White et al. (1990) and Illumina adaptor tails: ITS5-Illu-F and ITS2-Illu-R; ITS3-Illu-F and ITS4-Illu-R. PCRs were carried out in Thermal Cycler T100 (Bio-Rad, USA) under the programme: 95° – 10 min; 30 cycles each for 95° – 5 s, 50° – 20 s, 72° – 20 s; finally 72° – 5 min. PCR products (expected size of 400 bp) were checked on 1.2% agarose gels and purified with the suspension of magnetic beads Agencourt AMPure XP (Beckman Coulter, USA) in the ratio of 1:1.4 (v/v) and mixed in equimolar proportions for each mix. Second-stage PCR performed with 50 ng of mixed products and Nextera index primers (Illumina, USA), according to manufacturer’s instructions. Both PCR stages were conducted using Taq DNA polymerase (New England Biolabs, USA). Illumina libraries were sequenced on the MiSeq with MiSeq Reagent Kit v2 for 500 cycles (Illumina, USA).

The rbcL marker was sequenced on the 454 platform (GS Junior, Roche, Switzerland). The three-step PCR method was used for library preparation. All PCR stages were conducted using Phusion high fidelity DNA-polymerase (New England Biolabs, USA). The first-stage PCR was performed using a reaction mixture of a total volume of 20 µl: 4 μl 5× buffer for Phusion high fidelity DNA-polymerase, 250 µM dNTP (Thermo Scientific, USA), 0.2 μl Phusion high fidelity DNA-polymerase and 0.8 pM each primers (rbcLa-F and rbcLa-R, Hollingsworth et al. 2009).

The PCR conditions were as follows: 98 °C – 3 min; 7 cycles of 98 °C – 5 s, 50 °C – 30 s, and 72 °C – 30 s; finally 72 °C – 5 min. Amplification products were used without purification for the second round of PCR which was performed using primers rbcLa-454-F and rbcLa-454-R. PCR was performed using a reaction mixture of a total volume of 20 µl: 4 μl 5× buffer of Phusion high fidelity DNA-polymerase, 250 μM dNTP (Thermo Scientific, USA), 0.2 μl of Phusion high fidelity DNA-polymerase, 0.3 pM of each primer, 4 μl of amplification products of the previous stage. PCR conditions: 98 °C – 1 min; 18 cycles each for 98 °C – 5 s, 55 °C – 30 s and 72 °C – 30 s; finally 72 °C – 5 min. PCR products (expected size of 600 bp) were checked on 1.2% agarose gels and without purification used for third-stage PCR with primers MIDh-454-F and MIDh-454-R, containing barcodes (MIDx) and adapter sequences for sequencing on the 454 platform (GS Junior Systems, Roche, Switzerland). Adapter sequences are described in Gholami et al. (2012). Sample preparation was carried out with a protocol for the sequencing of amplicons Lib-A. Sequencing of the 454 platform (GS Junior, Roche, Switzerland) was performed according to the manufacturer’s instructions.

The matK marker was amplified using the primers matK_f (KIM3F) and matK_r (KIM1R) (Hollingsworth et al. 2009). A total of 50 ng DNA was used for a total volume of 25 μl reaction mix, containing 1 μM of each primer, 200 μM of each dNTP and 0.5 U Taq DNA polymerase (“New England Biolabs”, USA). PCRs were carried out in Thermal Cycler T100 (Bio-Rad, USA) under the programme 95 °C – 10 min; 30 cycles each for 95 °C – 5 s, 57 °C – 30 s, 72 °C – 30 s; finally 72 °C – 5 min. PCR products (expected size of 900 bp) were checked on 1.2% agarose gels and purified with the suspension of magnetic beads Agencourt AMPure XP (“Beckman Coulter”, USA) in the ratio of 1:1 (v/v). Sanger sequencing was performed on an Applied Biosystems 3730 DNA Analyzer using ABI PRISM BigDye Terminator v. 3.1 reagents. The reads were assembled using CLC Genomics Workbench ver. 10.0.1 (CLC Bio, Aarhus, Denmark; https://www.qiagenbioinformatics.com).

In addition to the 72 nrITS, 65 matK and 65 rbcL sequences generated in the present study, some published sequences (see Table 1) were taken from the INSDC (International Nucleotide Sequence Database Collaboration).

Sequences were aligned with MAFFT version 7 using default parameters (Katoh et al. 2017) and the alignment was adjusted manually using PhyDe (version 0.9971; Müller et al. 2010). Simple indel coding, according to Simmons and Ochoterena (2000) implemented in SeqState v.1.40 (Müller 2005), was applied to the ITS matrix which showed length mutational variation in contrast to the plastid DNA matrix. Phylogenetic relationships were reconstructed separately for the nuclear and plastid DNA datasets using maximum parsimony (MP), Bayesian inference (BI) and maximum likelihood (ML). The MP analyses were performed applying the parsimony ratchet (Nixon 1999) with PRAP v.2.0 (Müller 2004) in combination with PAUP v.4.0b10 (Swofford 2003), using 200 ratchet iterations with 25% of the positions randomly upweighted (weight = 2) during each replicate and 10 random addition cycles. Jackknife (JK) support values were calculated in PAUP with 10,000 jackknife replicates and the TBR branch swapping algorithm with 36.788% of characters deleted and one tree held during each replicate. The ML analyses were done with RaxML (Stamatakis 2014) in the version of RAxMLHPC v.8 on XSEDE on the CIPRES Science Gateway (Miller et al. 2010). Four partitions were separated in the nrITS dataset (three DNA: ITS1, 5.8s, ITS2, one binary), whereas the plastid DNA dataset with exclusively coding sequences was left unpartitioned. A rapid bootstrap analysis and search for the best-scoring ML tree in one single programme run with 100 bootstrap replicates was carried out using the general time-reversible (GTR) + Γ model as the standard predefined substitution model in RAxML. The BI analyses were performed with the MPI version of MrBayes (Ronquist et al. 2012) on the high-performance computing system of the Freie Universität Berlin. Instead of a priori testing of the optimal substitution model, sampling across the entire general time reversible (GTR) model space in the Bayesian MCMC analysis was applied (Huelsenbeck et al. 2004). Four partitions were separated in the nrITS dataset (as in ML) and two partitions in the plastid DNA dataset (matK, rbcL). Two simultaneous runs of four parallel chains each were performed for 3 × 10⁷ generations with a sample frequency of 1 tree per 2000 generations. Convergence of the runs was checked by making sure that the average standard deviation of split frequencies of the post-burn-in runs was below 0.01 and the effective sampling size (ESS) well above 200 in either run for all parameters. TreeGraph v.2 (Stöver and Müller 2010) was used to visualise the trees with statistical node support.

In total, 149 species of Scorzonerinae including Scorzonera s.l. (141 sp.), Koelpinia (3 sp.), Pterachaenia (1 sp.), Epilasia (3 sp.) and Tourneuxia (1 sp.) were carpologically studied. Carpological data for the genera Tragopogon and Geropogon were taken from a previous study by Sukhorukov and Nilova (2015).

Only marginal achenes of a capitulum were selected for the analysis. The cross-sections were made by hand in three topographical zones: in the basal third of the achene body, in the median third and in the apical third (or beak if present). The medium portion of the outer achenes is the most informative part, where all features are fully developed and was used for the carpological descriptions and phylogenetic reconstruction. Longitudinal sections were prepared in some cases to detect the peculiarities in the parenchymatous structures of the mesocarp. Prior to sectioning, the achenes were soaked in a mixture of ethyl alcohol, water and glycerol in equal proportions for a few days at 30–40^oC. The sections were stained with 2% carbol fuchsin, then with 0.4% picroindigocarmine and 100% ethanol (preparation of the reagents after Barykina et al. 2004).

The morphology of the achenes was documented using an Olympus SZ61 camera; the cross-sections were studied using Nikon Eclipse Ci microscopes and documented with a Nikon DS-Vi1 camera (Nikon Corporation, Japan) at the Department of Higher Plants (Moscow State University). The ultrasculpture of the achene surface was investigated using an SEM (JSM-6380 LA) at the Laboratory of Electron Microscopy Center at Moscow M.V. Lomonosov State University.

Describing carpological features, we applied the tripartite descriptive model (see, for example, Henning et al. 2018), where a feature comprises a character, which is, for its definition, decomposed into a “structure” (a physical component of an organism, for example, a carpophore) and a “property” (e.g. presence) and they were coded with binary or multiple unordered categorical states. Based on our carpological analysis, we selected and defined 16 achene characters potentially relevant for the reconstruction of character evolution. The character data matrix was built and analysed with Mesquite version 3.5 (Maddison and Maddison 2014; Mesquite Project Team 2014). Reconstruction of the character states at ancestral nodes was analysed with MP using the nrDNA tree as hypothesis of phylogenetic relationships and the ‘Trace character history’ and ‘Trace all characters’ tools as display methods. The character definitions and their state coding are provided under Results.

Phylogenetic reconstruction based on the nrITS region

The alignment of the nrITS region had a length of 707 characters; together with the coded indels, the matrix included a total of 832 characters, of which 422 were parsimony-informative. The MP analysis resulted in 697 most parsimonious trees (L = 2202, CI = 0.416, RI = 0.848, RC = 0.353, HI = 0.584). The 50% majority consensus tree was largely congruent in topology with the 50% majority consensus trees of the BI and ML analyses. Fig. 1 shows the BI cladogram with the MP jackknife (JK) support values above and the BI posterior probabilities (PP) and ML bootstrap (BS) support values below the branches.

The Scorzonerinae are resolved as monophyletic with strong statistical support (clade 1; JK = 97, PP= 1, BS = 100). The subtribe includes three major clades (1A, 1B, 1C), which received strong (1C; JK = 95, PP = 1, BS = 98) or full statistical support (1A, 1B). Their relationships to each other are unresolved. Clade 1A solely comprises the genus Tourneuxia. Clade 1B includes one part of Scorzonera s.l. We designate it as Gelasia clade, because it includes Scorzonera villosa, which provides the type of Gelasia Cass., being the oldest available generic name for members of this clade. Clade 1C includes the large remainder of the subtribe separated in two well-supported subclades. One (clade 1C1, JK = 95, PP = 1, BS = 99) comprises the genera Tragopogon, Geropogon and Epilasia; the other (clade 1C2, JK = 74, PP = 0.96, BS = 98) is a polytomy of four clades: (1) Scorzonera divaricata forming a clade of its own; (2) Pterachaenia with S. codringtonii as sister; (3) Koelpinia and a S. polyclada clade; (4) a large clade including both Scorzonera in the sense of its type S. humilis and Podospermum in one well supported subclade (JK = 76, PP = 1, BS = 98) and in a second moderately supported subclade (PP = 0.9, BS = 67) Pseudopodospermum and an extended Takhtajaniantha clade. Further details will be treated in the Discussion below.

Phylogenetic reconstruction based on the concatenated plastid DNA markers

The alignment of the two concatenated plastid DNA markers had a length of 1387 characters, of which 112 were parsimony-informative. The MP analysis resulted in 103 most parsimonious trees (L = 371, CI = 0.730, RI = 0.857, RC = 0.626, HI = 0.270). The 50% majority consensus tree was largely congruent in topology with the 50% majority consensus trees of the BI and ML analyses. Fig. 2 shows the BI cladogram with the MP jackknife (JK) support values above and the BI posterior probabilities (PP) and ML bootstrap (BS) support values below the branches.

The Scorzonerinae are resolved as monophyletic and Tourneuxia as sister to the remainder of the subtribe with strong support (clade 1; JK = 96, PP = 1, BS = 95). The latter clade 2 is a polytomy: clade 2A with almost full support includes the Gelasia clade; clade 2B with strong support only in MP includes Geropogon, Tragopogon and the Scorzonera divaricata clade; clade 2C with low support (JK = 65, PP = 0.97, BS = 61) includes Pterachaenia with S. codringtonii, the S. polyclada clade and Koelpinia; clade 2D includes with moderate support (JK = 75, PP = 0.99, BS = 73) Pseudopodospermum with Epilasia as sister; clade 2E includes a moderately supported (JK = 75, PP = 93, BS = 63) Takhtajaniantha clade and the moderately strong supported clade 2F includes both Scorzonera in the sense of the type and Podospermum.

Fewer of the deep nodes are resolved in the plastid DNA tree compared to the nrITS tree, but both reconstructions revealed largely the same major terminal clades. Statistically significant topological differences are few. They mainly concern the relationships of three clades: (1) Epilasia appears as sister to a clade combining Geropogon and Tragopogon in the nrITS tree, whereas as sister to Pseudopodospermum in the plastid DNA tree; (2) Scorzonera divaricata is sister to the Tragopogon-Geropogon clade in the plastid tree, but nested in a polytomy in the nrITS tree without the latter clade; (3) the Scorzonera purpurea-S. renzii clades are nested in the nrITS tree in a polytomy with S. angustifolia and the S. albicaulis clade, but in the plastid DNA tree, these two species are found in a polytomy with the Podospermum clade.

The capitula of the Scorzonerinae are, in principle, homocarpic. Minor differentiations corresponding to the centripetal development of the florets and thus also achenes in the Asteraceae capitula may, however, occur. The marginal achenes are considered as the most representative type. Achene wall anatomy and morphology are fully developed in the middle third of the achene as in all Asteraceae. Correspondingly, all features refer to the middle third of the marginal achenes.

The achene wall (Fig. 3) is composed of five segments corresponding to the five principal vascular bundles of the achene wall (plesiomorphic state in all Cichorieae and all Compositae) that may be barely noticeable. The principal vascular bundles of the achene wall can be seen often well below the principal ribs even if the latter are not clearly expressed. In the Scorzonerinae, each segment forms one main rib and, usually, two secondary ribs, the latter shared with the contiguous segments, resulting in a pattern of 10 ribs, either differentiated or not into more prominent principal and less prominent secondary ribs. The achene epidermis is the outer cell layer of the achene wall and can be glabrous, hairy or papillose. The achene surface can also exhibit emergences: stout cell conglomerations, hooked spines or tubercles with verrucose ribs consisting of pore parenchyma. Below the epidermis, the achene wall is composed of sclerenchymatous and parenchymatous layers of various arrangements and thickness. Sometimes, air cavities are present, resulting from rupture of thin-walled parenchyma. The endocarp is often obliterated. The seed coat is ± adjoining to the endocarp and consists of two or several layers, sometimes with the vestiges of vascular bundles. The endosperm is two-layered and the vertical embryo occupies almost all volume in the achene corpus.

The morphological characters of the pappus, as part of the diaspore, are also included in the carpological analysis.

The carpologial analysis across the subtribe revealed a variety of features, which were found to be diagnostic for species or groups of them. In the following, these carpological features are coded in characters (composed of a structure and a property) and states. In the character designation, structures and substructures are separated by colons and these are separated from the respective property by a semicolon.

1. Achene: carpopodium (formed by the tube-like protruding achene wall); presence

0: absent

1: present

Carpopodia as a protrusion of the basal achene wall surrounding the stipe-like structure in which the vascular bundles of the achene enter the receptacle, are known throughout the family and the tribe (Haque and Godward 1984). A particular conspicuous tube-like protrusion of the achene wall is present in some members of Scorzonera s.l. (entirely omitted by Haque and Godward 1984) and has been employed in the past as a diagnostic character to delimit taxa at infrageneric or generic rank (S. subg. Podospermum, S. sect. Purpurea and Pseudopodospermum).

2. Achene: carpopodium: border (by shape and surface texture) between achene corpus and carpopodium; presence

0: absent (Fig. 4A)

1: present (Fig. 4B, C)

We note here for the first time that the carpopodium is morphologically separated from the achene corpus by a border in Scorzonera subg. Podospermum and some members of S. subg. Pseudopodospermum

3. Achene epidermis: hairs; presence

0: none, epidermis glabrous

1: with soft multicellular eglandular hairs (Fig. 4D)

2: with drastically elongated, often slightly curved papillae looking like glandular hairs (Fig. 5A)

3: with cylindrical, not drastically elongated papillae or mamillae (Fig. 5B)

4: with retrorse acute papillae (Fig. 5C1, C2)

The outgrowths (hairs, papillae and mamillae), if present, originate from the achene epidermis. The hairs are multicellular in contrast to papillae (elongated cylindrical outgrowths) and mamillae (very small conical outgrowths) that are always unicellular. The character states 2 and 4 are peculiar for Koelpinia and Epilasia, respectively.

4. Achene surface: emergences; presence

0: absent

1: present

5. Achene surface: emergences; shape (Fig. 5)

0: stout and short conglomerations (Fig. 5D, E)

1: as hooked spines (Koelpinia: Fig. 5F)

2: as verrucose and undulate sculptures formed by pore parenchyma (Fig. 5G)

3: as spinules (Pterachaenia stewartii: Fig. 5H)

Emergences are distinguished from hairs and papillae as outgrowths of the achene wall originating from pericarp layers below the epidermis.

6. Achene wall: sclerenchyma; arrangement

0: Continuous sclerenchymatous layers (sheath) equal in thickness present

1: Discontinuous sclerenchymatous layers with a gap in the principal ribs

2: Continuous sclerenchymatous layers with a narrow or wider hunch on either side of the principal rib

3: Continuous or slightly discontinuous sclerenchyma with well-expressed invaginations (depressions).

Sclerenchyma as stabilising element develops in the achene wall above the vascular bundles into different patterns and formations as defined in the states coding.

7. Achene wall: sclerenchyma; orientation

0: Sclerenchyma entirely with parallel orientation to achene axis

1: Sclerenchyma differentiated: outer layer with parallel, inner with oblique to perpendicular orientation to achene axis.

2: Sclerenchyma only in emergences or rib areas perpendicular (or oblique) to achene axis.

8. Achene wall: parenchyma; arrangement (Fig. 6)

0: Present as subepidermal continuous layer(s)

1: Continuous parenchyma layers above and below the sclerenchyma

2: Insular in principal ribs below sclerenchyma and in secondary ribs above sclerenchyma

3: Only insular in principal ribs below sclerenchyma

4: Insular above sclerenchymatous invaginations and below sclerenchyma in the principal ribs

5: Only insular above sclerenchymatous invaginations

6: Absent

7: Parenchyma continuous above sclerenchyma and below sclerenchyma insular in principal ribs

8: Only insular in second ribs above sclerenchyma

Besides sclerenchyma, the main element of the achene wall is parenchyma, but often the parenchyma elements are left behind in quantity by the sclerenchyma. The distribution of the parenchyma elements in the achene wall is taxon specific and several different arrangements have been found as outlined in the character state coding. The various ways the parenchyma can be arranged in the achene wall is schematically shown in Fig. 6. As in the other cases, we have refrained from speculations about evolutionary pathways for these arrangements and have coded them as unordered states.

9. Achene wall: parenchyma; differentiation

0: Parenchyma only as subepidermal mechanical parenchyma

1: Subepidermal collenchyma-like layers differentiated besides parenchyma

2: Parenchyma of thin-walled cells only

3: Parenchyma of two types present, mechanical parenchyma and such with thin-walled cells

The unusual collenchyma-like layers are built of thin-walled cells with prominent intercellular spaces of triangular shape. The parenchyma in character states 0 and 2–3 lack prominent intercellular spaces.

10. Achene wall: air cavities; occurrence

0: Absent

1: Present

11. Achene wall; ribbing pattern

0: Each segment with a principal rib and 2 secondary ribs, the latter shared with the contiguous segments, achene middle third, thus with 5 principal and 5 secondary ribs

1: Each segment with a principal rib only, achene middle third, thus with 5 ribs

2: No distinct ribs developed, middle third ± terete (roundish)

3: Two or more ribs enlarged to wings

12. Achene wall: tannins; presence

0: Absent

1: Present in cell protoplast

2: Present in cell walls only

13. Achene beak; presence

0: Absent

1: Present

The beak is defined as the more or less abrupt attenuation of the achene apex into a stipe-like structure carrying the pappus disc.

14. Pappus; presence

0: Absent

1: Present

15. Pappus; structure

0: Of entirely softly plumose bristles (with long soft fimbriae all along the bristle)

1: Of entirely softly plumose bristles and scabrid awns

2: Of bristles softly plumose in lower and scabrid in upper half or third (at least 5–10 longer bristles have unequal lateral fimbriae)

4: Bristles scabrid completely or for the most part

The setaceous pappus in the Cichorieae consists of bristles with lateral projection (= fimbriae) not or little exceeding the diameter of the bristles, which make the bristles rough, thus “scabrid”. Alternatively, the lateral projection may be many times longer than the bristle diameter and the bristles thus featherlike or “plumose”. Plumose pappi in the Cichorieae have evolved as three different types, each specific for one subtribe (Kilian et al. 2009a). The plumose bristles in the Scorzonerinae are characterised by soft and often intertwining fimbriae pointing in all directions and consisting of a row of flattened cells. Taxon specific variation of the pappus structure concerns the composition of the pappus by plumose and scabrid elements or the differentiation of the fimbriae development in one and the same pappus element.

16. Pappus; colour

0: white

1: dirty white

2: fulvous (at least in basal part)

3: yellowish

4: grey or blackish

The matrix of the 16 carpological characters is given in Table 2. Representative achene cross sections photographs and schemes representative for the clades are presented in Figs 7–14. The detailed carpological descriptions of the principal terminal clades are presented in the concluding part “Taxonomy”, below. Here, we summarise the anatomical features for the individual clades.

Tourneuxia clade

Fig. 7A, B

Achenes without carpopodium, with two ribs elongated into small wings; achene epidermis glabrous; emergences absent; subepidermal parenchyma almost continuous, more prominent in the winged areas and much thinner (1–3 layers) between them, of thin-walled cells only; sclerenchyma continuous, its fibres of parallel orientation, sometimes slightly obliquely orientated in the region of the wings; no air cavities; tannins absent.

Gelasia clade

Fig. 7C, D

Achenes without carpopodium; achene epidermis often densely covered with soft multicellular hairs making a long woolly indumentum; emergences mostly absent; subepidermal parenchyma usually insular, of thin-walled cells only; sclerenchyma continuous with an invagination on either side of the principal rib, parallel; no air cavities; tannins absent.

Geropogon clade

Fig. 8A, B

Achenes without carpopodium; achene epidermis with papillae, emergences present; subepidermal parenchyma continuous of thin-walled cells only; sclerenchyma continuous, parallel; air cavities present; tannins present.

Tragopogon clade

Achenes without capopodium; achene epidermis usually with papillae; emergences absent or present; subepidermal parenchyma continuous, of thick-walled or thick- and thin-walled cells; sclerenchyma continuous, parallel; air cavities usually present; tannins absent.

Epilasia clade

Fig. 8C, D

Achenes with carpopodium; achene wall with hair-like papillae; emergences absent; subepidermal parenchyma continuous, of thick-walled cells only; sclerenchyma continuous, parallel; air cavities absent; tannins present.

Scorzonera divaricata clade

Fig. 9A, B

Achenes without carpopodium; achene epidermis with papillae; emergences absent; subepidermal parenchyma of both thin- and thick-walled cells; sclerenchyma continuous, parallel; air cavities absent; tannins present.

Pterachaenia clade

Fig. 9C, D

Achenes without carpopodium; with 2–3 elongated ribs forming wings (Pterachaenia stewartii) or without (Pterachaenia codringtonii); achene epidermis with papillae or glabrous; emergences absent; subepidermal parenchyma discontinuous, located above sclerenchyma between the principal ribs and below sclerenchyma in the rib areas or only insular above sclerenchyma, of both thin- and thick-walled cells; sclerenchyma continuous, parallel; air cavities absent; tannins absent.

Scorzonera polyclada clade

Fig. 10A, B

Achenes without carpopodium; achene epidermis glabrous; emergences absent; subepidermal parenchyma continuous; sclerenchyma usually continuous (sometimes irregularly discontinuous), with an invagination on either side of the principal rib, parallel; air cavities absent; tannins usually absent (rarely present in cell wall only).

Koelpinia clade

Fig. 10C, D

Achenes without carpopodium; achene epidermis glandular-like (elongated) papillae (almost glabrous in K. tenuissima); emergences present; subepidermal parenchyma continuous, of thick-walled cells only; sclerenchyma continuous, parallel (but perpendicular in emergences); air cavities absent; tannins absent.

Takhtajaniantha clade

Fig. 11A–D

Achenes without carpopodium; achene epidermis with papillae or soft multicellular hairs; emergence absent or present; subepidermal parenchyma continuous; air cavities absent; tannins absent. Two types represented: (1) with parenchyma of both thin- and thick-walled cells and sclerenchyma continuous with a narrow or wider hunch on either side of the principal ribs (Takhtajaniantha type: Scorzonera austriaca, S. austriaca subsp. curvata, S. austriaca var. verrucosa, S. austriaca var. tenuifolia, S. crispa, S. ikonnikovii, S. mongolica, S. pusilla; Fig. 11A, B); (2) with parenchyma of thick-walled cells only and sclerenchyma either continuous or discontinuous forming 5 bundles, without hunches (Pseudodivaricata type: S. capito, S. pseudodivaricata, S. tau-saghyz, S. veresczaginii; Fig. 11C, D).

Pseudopodospermum clade

Fig. 12A–F

Achenes with or without carpopodium; achene epidermis usually with papillae; emergences usually present; subepidermal parenchyma usually continuous, often reduced to one layer or few layers, of thin-walled cells only or of both thick- and thin-walled cells; air cavities absent or present in the principal ribs; sclerenchyma with parallel fibres, of three types (1) with a narrow or wider hunch on either side of the principal ribs (Brevicaulis type: S. brevicaulis, S. crocifolia, S. hispanica var. asphodeloides, S. phaeopappa, S. syriaca; Fig. 12A, B), (2) discontinuous with a gap in the principal ribs (Calyculata type: S. calyculata, S. crispatula, S. davisii, S. hissarica, S. inaequiscapa, S. incisa, S. libanotica, S. ovata, S. papposa, S. reverchonii, S. troodea, S. violacea; Fig. 12C, D) or (3) continuous and equal in thickness (Pseudopospermum type: S. boetica, S. chantavica, S. elata, S. hispanica, S. idaea, S. inconspicua, S. gracilis, S. lamellata, S. leptophylla, S. mollis, S. mollis var. platyphylla, S. pachycephala, S. pubescens, S. raddeana, S. rawii, S. stricta, S. suberosa, S. szowitzii, S. turcomanica, S. semicana, S. undulata; Fig. 12E, F.

Scorzonera s.str. clade

Fig. 13A, B

Achenes without carpopodium; achene epidermis with papillae; emergences present (S. aristata) or absent; parenchyma insular in principal ribs below sclerenchyma and in secondary ribs above sclerenchyma, of thin-wall cells only; sclerenchyma continuous, parallel; air cavities absent; tannins absent.

Scorzonera purpurea clade

Fig. 13C, D

Achenes with carpopodium; achene epidermis with papillae; emergences mostly present; parenchyma represented only by one layer or few layers above and below sclerenchyma, of both thin- and thick-walled cells; sclerenchyma continuous, parallel; air cavities absent; tannins absent.

Scorzonera albicaulis clade

Fig. 14A, B

Achenes without carpopodium; achene epidermis with papillae; emergences absent; sclerenchyma with parallel fibres; air cavities usually absent; tannins absent. Two types represented: (1) subepidermal parenchyma represented by collenchymatous cells and sclerenchyma with a gap in the principal ribs (Piptopogon type: Scorzonera angustifolia, S. graminifolia, S. baldschuanica, S. bracteosa, S. crassicaulis, S. petrovii, S. tragopogonoides; Fig. 14A, B) and (2) subepidermal parenchyma with thin-walled cells and sclerenchyma continuous (Albicaulis type: Scorzonera acanthoclada, S. albicaulis, S. franchetii, S. iliensis, S. racemosa, S. rupicola, S. transiliensis, S. turkestanica, similar to Scorzonera s.str.; Fig. 13A, B).

Podospermum clade

Fig. 14C, D

Achenes with carpopodium; achene epidermis glabrous or with multicellular eglandular hairs; emergences mostly absent; subepidermal parenchyma continuous of thin-walled cells or both of thin- and thick-walled cells; sclerenchyma continuous, diversely orientated (outer layers orientated parallel to the achene axis, the inner ones obliquely or perpendicularly orientated); air cavities absent; tannins absent.

The results of the ancestral character reconstruction, based on the nrITS tree, are presented in Suppl. material 1: Fig. S1. For further details, see the corresponding section in the Discussion.