African Mountain Thistles: Three New Genera in the Carduus-Cirsium Group

The floras on the highest mountains in tropical eastern Africa are among the most unique floras in the world. Despite the exceptionally high concentration of endemic species, these floras remain understudied from an evolutionary point of view. In this study, we focus on the Carduus-Cirsium group (subtribe Carduinae) to unravel the evolutionary relationships of the species endemic to the tropical Afromontane and Afroalpine floras, aiming to improve the systematics of the group. We applied the Hyb-Seq approach using the Compositae1061 probe set on 190 samples (159 species), encompassing representatives of all genera of Carduinae. We used two recently developed pipelines that enabled the processing of raw sequence reads, identification of paralogous sequences and segregation into orthologous alignments. After the implementation of a missing data filter, we retained sequences from 986 nuclear loci and 177 plastid regions. Phylogenomic analyses were conducted using both concatenated and summary-coalescence methods. The resulting phylogenies were highly resolved and revealed three distinct evolutionary lineages consisting of the African species traditionally referred to as Carduus and Cirsium. Consequently, we propose the three new genera Afrocarduus, Afrocirsium and Nuriaea; the latter did notably not belong to the Carduus-Cirsium group. We detected some incongruences between the phylogenies based on concatenation vs. coalescence and on nuclear vs. plastid datasets, likely attributable to incomplete lineage sorting and/or hybridization.


Introduction
The current biodiversity crisis entails numerous consequences and unprecedented challenges [1,2], emphasizing the urgent need to enhance our systematic understanding of biodiversity, as it serves as a crucial tool for its conservation [3].Despite hosting multiple biodiversity hotspots, the least studied areas of the world, from an evolutionary perspective, are located in the tropics [4,5].It is particularly urgent to fill such a knowledge gap given the disproportionately high extinction rates experienced by tropical regions [6] and the significant anthropogenic pressures they face, such as extensive land-use changes [7].
The Afromontane and Afroalpine archipelagos host unique tropical floras, which are part of the Afrotemperate biome [8,9].These tropical Afrotemperate flora thrive in humid and temperate isolated refuges on the highest African mountains, which are located primarily in the eastern part of the continent [10].Both the Afromontane flora (below the treeline) and the Afroalpine flora (above the treeline) harbor exceptionally high concentrations of endemic species (c.75%, [11,12]).Most of the research on plant evolution has focused on the Afroalpine flora (reviewed in Brochmann et al., 2021 [13]), leaving the Afromontane flora understudied.To address this knowledge gap, our study aims to investigate the Carduus-Cirsium group (subtribe Carduinae), which includes species that occur both below and above the treeline and are mostly endemic to tropical eastern Africa.
Despite recent research efforts [19,20], Cirsium and Carduus remain the most problematic genera.The recent taxonomic proposal for the Carduus-Cirsium group (20) split Cirsium into four genera, accepting Cassini's subgenus Lophiolepis at the generic level, reinstating the genus Epitrachys (DC.Ex Duby) K.Koch and describing the hybrid genus xLophiocirsium Del Guacchio, Bureš, Iamonico & P. Caputo, mainly based on the weakly supported molecular results from Ackerfield et al. (2020) [19].However, both studies lacked sufficient sampling across the Carduus-Cirsium group.For example, Afromontane Cirsium species were not included despite their distinct morphological differences from all other species in Cirsium.
Carduus, the second most speciose genus, is monophyletic only when excluding subgenus Afrocarduus Kazmi [29] from the phylogenetic analyses [17,19,25].Subgenus Afrocarduus is endemic to the mountains of tropical eastern Africa and is a good example of the adaptation, radiation and evolution of montane and alpine species.It includes both widespread taxa growing across several mountain regions and narrow endemics restricted to one massif, all of them between 1600 and 4600 m.a.s.l.[29][30][31].Afrocarduus species exhibit a combination of morphological and karyological characters typical of both Carduus and Cirsium.Species delimitation within this group is highly controversial.The two main taxonomic treatments differ conspicuously in the number of accepted species: 22 in Kazmi (1963) [29] and only 9 in Jeffrey (1968) [31].The controversies surrounding the delineation of Cirsium as well as the unresolved relationships of the Afromontane species underscore the need for a new approach to resolve the phylogeny and systematics of the Carduus-Cirsium group.Previous studies suffered from limited taxon and gene sampling, resulting in poorly resolved phylogenies.Next-Generation Sequencing (NGS) methods offer improved resolution in phylogenetic reconstructions at all taxonomic levels [32][33][34][35].Here, we compiled a comprehensive sampling of all the genera in the subtribe and applied a Hyb-Seq approach using a probe set targeting exons of 1061 orthologous loci developed for Compositae [36], which also enabled the retrieval of complete plastid genome sequences.The three main objectives of our study are to (1) infer the first phylogeny of the subtribe Carduinae that includes a comprehensive taxonomical and biogeographical sampling; (2) establish a delimitation of the genera with a particular focus on the Carduus-Cirsium group; and (3) unravel the evolutionary origins of the Afromontane and Afroalpine species.

Target Loci Recovery
Out of 1064 loci of the target reference, 918 were recovered with a mean coverage of the retained contigs of 45.38.The customized reference after splitting the paralogous sequences into orthologous alignments comprised a total of 1407 loci, of which 507 alignments were generated after splitting the paralogous ones.After mapping and filtering by missing data and sample presence, 986 loci were retained, of which 775 originally contained only orthologous sequences, whereas 211 originated from the splitting of paralogous sequences into orthologous alignments.Sequence divergence values between 8.6 and 22.7% were considered to indicate paralogy (Figure S1).The average alignment length was 250 bp (88-719), the average parsimony site length was 53 bp , the average variable site length was 76 bp  and the average missing data was 3.69% (0-65%).
For the concatenated analysis, the nuclear and plastid supermatrix had a length of 247,251 bp and 117,823 bp, respectively.The coalescence analysis was based on 986 alignments, and a gene tree was calculated for each of them.For the plastid data, we retrieved a total of 183 regions, of which we retained 177 after filtering for missing data and sample presence: the final plastid dataset included 30 tRNAs, 4 rRNAs, 79 coding genes (including 13 introns) and 64 intergenic regions.

Nuclear Dataset
The concatenated and coalescence approaches yielded phylogenies (Figures 1 and 2) that were highly congruent concerning the relationships between the main clades that fell outside the Carduus-Cirsium group, whereas we found several differences within this group (Figure 3).Both phylogenies retrieved subtribe Carduinae as a strongly supported monophyletic group (Figure 1, TBE = 1, BS = 100; Figure 2, LPP = 1).The genus Carduus was monophyletic in both analyses (Figure 1, BS = 92, TBE = 0.97, LPP = 1) only when subg.Afrocarduus was excluded.The genus Cirsium was monophyletic in the concatenated analysis when the African species were excluded.In the coalescence analysis, the African species fell outside Cirsium, and the non-African Cirsium species plus Picnomon formed a monophyletic group.The African species traditionally referred to as Cirsium dender I.Friis and C. englerianum O.Hoffm.formed a monophyletic group (Figure 1, BS = 100, TBE = 1; Figure 2, LPP = 1) outside the Carduus-Cirsium clade and as a sister to a clade formed by Galactites and Lamyropsis (Figure 1, BS = 100, TBE = 1; Figure 2, LPP = 1).The remaining tropical eastern African species traditionally referred to as Cirsium formed a monophyletic group (Figure 1, BS = 100, TBE = 1; Figure 2, LPP = 1) as a sister to the clade corresponding to subg.Afrocarduus (Figure 1, BS = 100, TBE = 1; Figure 2, LPP = 1).In the concatenated phylogeny (Figure 1), this clade (east African Cirsium + subg.Afrocarduus) was a sister to Picnomon+Notobasis+non-African Cirsium, whereas in the coalescence phylogeny (Figure 2), it was a sister to the main clade constituted by all the other lineages of the Carduus-Cirsium group.

Plastid Dataset
The maximum-likelihood tree inferred from the plastid supermatrix (Figure S2) yielded a highly supported phylogeny except for some shallow nodes within Cirsium and Carduus.The genera Cynara, Galactites, Lamyropsis and Ptilostemon were recovered as monophyletic with high support (all BS = 100, TBE = 1), as was the Carduus-Cirsium group (BS = 100, TBE = 1).Cirsium dender and C. englerianum were recovered outside the Carduus-Cirsium group, as in the nuclear analyses (Figures 1 and 2).The genus Cirsium was not recovered as monophyletic; it was split into four clades that were nested with other genera of the Carduus-Cirsium group.African Carduus were recovered as monophyletic (BS = 97, TBE = 0.98) and a sister (BS = 100, TBE = 1) to a clade formed by two species of African Cirsium (BS = 100, TBE = 1).In contrast, Cirsium straminispinum C.Jeffrey was recovered within one of the Cirsium clades.Silybum (BS = 100, TBE = 1) and Carduus (BS = 100, TBE = 1) were recovered as monophyletic when African Carduus were excluded.Tyrimnus (BS = 100, TBE = 1) was recovered as a sister to the latter, as in the two nuclear-based phylogenies (Figures 1 and 2).Notobasis and Picnomon were recovered as successive sister lineages to one of the Cirsium clades, in both cases with full support (BS = 100, TBE = 1).The genus Picnomon was recovered as a sister to Notobasis in the concatenated tree, but with low support (Figure 1, BS = 65, TBE = 0.65), forming a lineage that was a sister to Cirsium s. str.In contrast, in the coalescence phylogeny, Picnomon was nested within the genus Cirsium with high support (Figure 2 2, LPP = 1).In the concatenated phylogeny (Figure 1), this clade (east African Cirsium + subg.Afrocarduus) was a sister to Picno-mon+Notobasis+non-African Cirsium, whereas in the coalescence phylogeny (Figure 2), it was a sister to the main clade constituted by all the other lineages of the Carduus-Cirsium group.
The genus Picnomon was recovered as a sister to Notobasis in the concatenated tree, but with low support (Figure 1, BS = 65, TBE = 0.65), forming a lineage that was a sister to Cirsium s. str.In contrast, in the coalescence phylogeny, Picnomon was nested within the genus Cirsium with high support (Figure 2

Taxon Sampling
A total of 190 samples (from both herbarium specimens and specific field campaigns) of 159 species representing all the genera of subtribe Carduinae as defined by Herrando-Moraira et al. (2019) [32] were selected for our study (Table S1).Samples of Cirsium and Carduus were chosen to cover both their taxonomic diversity and subcosmopolitan distribution.Especially for the Afromontane and Afroalpine taxa, we sampled broadly to cover morphological heterogeneity and possible misleading taxonomical determinations.We also included two species of Silybum; the monotypic genera Notobasis, Picnomon and Tyrimnus; and two species of each of the other genera of Carduinae: Cynara L., Galactites Moench, Lamyropsis (Kharadze) Dittrich and Ptilostemon Cass.Finally, four species of Onopordinae (sister to Carduinae) were included as outgroup taxa based on previous phylogenetic results at the tribal level [32].For eleven species, we used the raw reads previously obtained in a latter study, whereas all other sequences were newly generated either from herbarium specimens or fresh material collected in the field and preserved in silica gel.

DNA Extraction, Library, Capture and Sequencing
For each sample, 1-50 mg of dried plant material was ground.DNA was extracted using an E.N.Z.A ® SP Plant DNA Kit (Omega Bio-Tek Inc., Norcross, GA, USA) following the manufacturer's instructions.DNA was measured using a Qubit™ Flex Fluorometer (Thermo Scientific, Waltham, MA, USA) and sheared (0.2-1 µg in 50 µL) using a Qsonica Q800R3 Sonicator (Qsonica LLC, Newtown, CT, USA) at 20% amplitude for 45 s to 10 min at 4 • C to obtain fragments of 300-400 bp.Fragment length was checked with electrophoresis in 1.2% agarose gels.Libraries were prepared using an NEBNext Ultra II DNA Library Prep Kit for Illumina ® (New England Biolabs, Ipswich, MA, USA) from 25 to 45 µL of sonicated DNA using half of the volumes stipulated by the manufacturer and fifteen cycles of PCR amplification.Libraries were barcoded with single or dual index primers, NEBNext ® Multiplex Oligos for Illumina ® .Libraries were quantified with Qubit and those with more than 17 ng of DNA were pooled to a max. of 10 samples and 2 µg of total DNA (around 250 ng for each sample).Subsequently, pools were evaporated or filled with water to 7 µL of volume to execute the target-enrichment protocol of [36] using a Microarray MyBaits COS kit (Daicel Arbor Biosciences, Ann Arbor, MI, USA).To sequence plastid DNA, 40% of the DNA from the libraries previous to the target enrichment step was added to each pool.Pools were sequenced (PE 150 bp) using HiSeq 2500 and HiSeq X.For some samples, the DNA was extracted and sent to Daicel Arbor Biosciences for library construction, target-enrichment capture and posterior sequencing on an Illumina NovaSeq (PE 150 bp).

Molecular Data Processing
HybPhyloMaker [37] scripts available at https://github.com/tomas-fer/HybPhyloMaker(accessed on 4 November 2021), a bioinformatic pipeline specially developed to process Hyb-Seq data, were used to process raw sequence reads and conduct phylogenetic inference analyses, in combination with ParalogWizard [38].ParalogWizard is a workflow developed to detect paralogous sequences and separate them into different alignments based on sequence similarity.Thus, this procedure allows for obtaining two or more orthologous alignments from an alignment that included paralogous sequences.This approach avoids losing potential informative sites in paralogous loci and reduces gene tree discordance in recent and rapidly radiated groups [39].
Raw reads were cleaned using Trimmomatic v.0.32 [40] as implemented in HybPhylo-Maker.To obtain the targeted nuclear data, the Compositae1061 probe set [36] was used for initial read mapping using BWA [41] and SPAdes scripts [42] as implemented in Par-alogWizard.Pairwise sequence divergence was calculated to identify paralogues, resulting in two peaks: the first represented putative allelic variation, and the second represented highly divergent sequences corresponding to putative paralogues.The divergence value of the second peak was used as the threshold value to identify putative paralogous sequences and separate them into orthologous matrices, which were aligned using MAFFT v.7.029 [43].The obtained alignments were processed with HybPhyloMaker (scripts 5 and 5b) to reduce missing data.We removed sequences missing more than 70% of the total locus length and loci for which less than 75% of the samples were represented.
To obtain plastid sequences, we created a reference file from the complete plastome sequence of Cirsium arvense (NCBI Accession number: KY562583) including a total of 201 regions: 86 genes (coding regions + introns), 8 rRNAs, 37 tRNAs and 70 intergenic regions extracted using Artemis v.18.2 [44].Quality-trimmed reads were mapped using this reference and the BWA method as implemented in HybPhyloMaker.The subsequent match of the mapped sequences with the probes, alignments and processing of the missing data was carried out using HybPhyloMaker with the same criteria as for the nuclear data (<70% of missing data and a >75% sample presence per gene).

Phylogenetic Analyses
We used both concatenation and coalescence approaches to infer phylogenetic relationships.In the concatenation approach, a nuclear-based phylogenetic tree was inferred based on a single supermatrix of all nuclear loci retrieved; the same process was repeated with all plastid regions combined in a second supermatrix of all chloroplastic regions retrieved.The coalescence approach to generate a species tree was only applied to the nuclear dataset since all chloroplast regions are assumed to be linked.
With each supermatrix, nuclear and chloroplast, phylogenetic inference analyses under maximum likelihood were carried out using RaxML-NG [45] with 20 independent tree searches and applying the best-fit model for each partition (one for each locus) previously determined using Modeltest-NG [46].To measure branch support, Felsenstein's Bootstrap BS [47] and the Transfer Bootstrap Expectation TBE [48] were calculated using RaxML-NG and applying the bootstopping criterion with the default to determine the sufficient number of replicates, cf.[49,50].We included TBE in addition to BS because it has been reported that the former is more suitable for studies with large datasets.In TBE, the presence of inferred branches in bootstrap trees is measured with a gradual 'transfer' distance, while in BS, it is based on binary presence/absence [48].As a result, when using a large number of taxa, it becomes harder to resample clades exactly as in the original tree, causing an underestimation in large clades and obtaining significantly lower values of BS at deep nodes [51].For both metrics, the value of 70% was chosen as the threshold indicating supported branches.
For the coalescence approach, nuclear gene trees were inferred using RaxML v.8.2.12 [52] with a bootstrap resampling of 100 replicates.Based on these gene trees, a species tree was obtained using ASTRAL-III v.5.7.8 [53].Support values were calculated using local posterior probabilities LPP [54], and branches were considered well-supported at LPP ≥ 0.95.To detect the degree of incongruence between gene trees, a quartet-based method analysis was run with the −t 8 option of ASTRAL-III, which allows for identifying the percentage of genes that support alternative topologies for each node [55].

Morphological Examination
Both before and after the phylogenetic analyses, a morphological study of the genera was undertaken.We studied a total of 74 herbarium sheets from herbaria (Institut Botànic de Barcelona (BC), Meise Botanic Garden (BR), Muséum National d'Histoire Naturelle (P) and Naturalis Biodiversity Center (WAG)), and material collected in the field.The complete list of specimens examined is provided in the description of the new genera.The characters we examined (Table S2) were those considered relevant by Kazmi (1963) [29].

Utility of Hyb-Seq and Incongruence between the Phylogenies
Our analyses yielded highly supported phylogenetic reconstructions, reaffirming the effectiveness of NGS combined with the target-capture methodology proposed by Mandel et al. (2014) [36] in resolving relationships in Compositae at different taxonomical levels [32].Based on the extensive nuclear dataset comprising 986 loci, we successfully constructed a phylogeny for Carduinae with strongly supported groups.
Concerning the relationships among the main clades, we found some incongruence between the nuclear concatenated and coalescence phylogenies (Figures 1 and 2).In the concatenated tree, the East African clade was a sister to a clade consisting of the remaining Cirsium species plus Notobasis and Picnomon, whereas in the coalescence tree, the East African clade was a sister to the Carduus-Cirsium clade (Figure 2).Such inconsistency may be attributable to an incongruent phylogenetic signal among genes due to incomplete lineage sorting (ILS), which has been shown to be problematic in concatenated approaches [56,57].This explanation is supported by our quartet support analysis (Figure 2), which indicated nearly equal proportions of genes supporting alternative topologies for the nodes concerned.Such results have been suggested to indicate high levels of ILS [55].
The plastome analysis yielded a highly supported tree also showing the monophyly of the Carduus-Cirsium group, and suggests that the African species of this group represent three independent evolutionary lineages (except for Cirsium straminispinum; see details below).The main incongruence between the plastid and nuclear trees was found within the Carduus-Cirsium group, concerning Cirsium taxa (Figure 3).In the nuclear trees, the non-African Cirsium species were either recovered as a separate clade (Figure 1) or nested with Picnomon (Figure 2).In contrast, the plastid tree split Cirsium into four clades: one close to Picnomon and Notobasis and three forming a graded clade in which Carduus + Tyrimnus are nested (Figure 3).Such incongruences may result from several non-exclusive factors that promote unstable phylogenetic relationships, such as rapid radiations, ILS and hybridization [58][59][60].Indeed, hybridization or introgression have been reported multiple times in the evolutionary history of the tribe Cardueae, e.g., [61].

The African Carduus and Cirsium
The monophyly we inferred for the African species traditionally referred to as Carduus (Figure 1, Figure 2 and Figure S2) is consistent with previous morphological [17,29] and molecular studies [19,25].The main difference between Eurasian and African Carduus (subg.Afrocarduus) is found in the morphology of the achene.Häffner (2000) [17] was the first to point out that these taxa could constitute a lineage more related to Cirsium than to Carduus based on the shape of the dorsal corolla lobe epidermis (straight as in Cirsium, not undulate as in Carduus), indument of the stamen filaments (glabrous as in Cirsium, pilose as in Carduus) and characters of the achene pericarp (with four blunt ribs and persisting at maturity as in Cirsium, not with 10-15 longitudinal stripes and disintegrating at maturity as in Carduus).Their chromosome numbers are also similar: Carduus usually has n = 9, and both Carduus subg.Afrocarduus and Cirsium have n = 16-17 (17).Our examination of 74 herbarium specimens confirmed all reported morphological differences.The reason for the traditional placement of these African species in Carduus is that Carduus and subg.Afrocarduus always lack plumose pappus, which is found in Cirsium s. str.and African Cirsium.Our results thus demonstrate that the African Carduus form a distinct evolutionary lineage that does not belong neither to Cirsium nor to Carduus, that it can be diagnosed using several morphological characters and that it is endemic to the tropical mountains of eastern Africa.Thus, a new classification that accurately represents the phylogenetic history of the group is needed.Here, we propose a new genus using the name provided by Kazmi (1963) [29] for the subgenus: Afrocarduus (Kazmi) Garcia-Jacas, Moreyra & Susanna, gen.et stat.nov.Concerning the number of species-nine (Jeffrey, 1968 [31]) vs. 22 (Kazmi, 1963 [29])-our results and our ongoing research (Moreyra et al., unpubl.data) seem to suggest a number close to that suggested by Kazmi (1963) [29].For example, Carduus kikuyorum R.E.Fr. was considered a subspecies of Carduus nyassanus R.E.Fr.by Jeffrey (1968), but it should be considered an independent species as proposed by Kazmi (1963) [29], since they are independent evolutionary lineages, according to our phylogenies (Figures 1 and 2).A complete taxonomic treatment for Afrocarduus is under preparation and will be published in a separate work.
Three of the African species traditionally referred to as Cirsium (Figures 1 and 2; C. buchwaldii O.Hoffm., C. schimperi (Vatke) C.Jeffrey and C. straminispinum) were recovered together as a clade sister to Afrocarduus with the nuclear dataset.Cirsium straminispinum was recovered within the Cirsium clade in the plastome analysis (Figure S2), and we consider it likely that this may be a signal of a hybrid origin of this species [25].These three species have the typical plumose pappus of Cirsium [14], but they also show distinctive characters such as phyllaries with well-developed pectinate appendages, which are absent in Cirsium and in all other genera in the Carduus-Cirsium group.We are aware of the weakness of suggesting a new taxon based on a single morphological character, especially given the difficulties to differentiate groups in the subtribe Carduinae [18].However, we consider the presence of a unique diagnostic morphological character combined with strong evidence of being an independent evolutionary lineage to be sufficient to propose a new genus, Afrocirsium Calleja, Garcia-Jacas, Moreyra & Susanna, gen.nov., that is a sister to Afrocarduus in all the analyses.
Finally, two Ethiopian species traditionally referred to as Cirsium, C. dender and C. englerianum were recovered as a monophyletic group falling outside the Carduus-Cirsium clade, as a sister to Galactites and Lamyropsis (Figures 1 and 2) or a sister to Galactites (Figure S2).These species share the plumose pappus with Cirsium s. str.and Afrocirsium.However, as stressed by Friis (1975) [62], the two species differ conspicuously from Cirsium/Afrocirsium by their large capitula (4-7 cm) that resemble those of Cynara [63]; a strong (>2 mm width) and long thorn (>30 mm) on their basal leaf lobes; and their large size (2-5 m).We therefore propose a new genus, Nuriaea (Friis) Susanna, Calleja & Moreyra, comb.nov., to accommodate these two species.

The Carduus-Cirsium Group
The Carduus-Cirsium group was recovered as monophyletic, and after segregating the new genera Afrocarduus and Afrocirsium, it contains eight genera: Afrocarduus, Afrocirsium, Carduus, Cirsium, Notobasis, Picnomon, Silybum and Tyrimnus (Figures 1 and 2).The main incongruence between the coalescence and concatenation trees concerns the positions of the monotypic genera Picnomon and Notobasis.Notobasis was recovered as a sister to Picnomon (Figure 1), to Carduus + Tyrimnus + Silybum (Figure 2) or to Picnomon + Cirsium (Figure S2).These inconsistencies could result from ILS and/or hybridization (as discussed above).In all analyses, Notobasis was recovered as an independent lineage with a long branch length, and we, therefore, suggest maintaining its generic status.We also tentatively favor maintaining Picnomon as a distinct genus because (1) it was recovered as an independent lineage in the nuclear and plastid trees obtained under concatenation; (2) in the coalescence tree, it was instead recovered as a sister to one of the two main clades of Cirsium, but both the node leading to Picnomon and the preceding node had similar proportions of genes supporting alternative topologies (Figure 2).A final decision to confirm the appropriateness of this choice should await further research including more individuals.
The genus Cirsium.The status of Cirsium (for all non-African species) as a single entity has often been questioned, and the need for further research using more accurate methods to make taxonomical decisions has been pointed out [19].Recently, the authors of [20] split Cirsium into four genera (Cirsium s. str., Lophiolepis, Epitrachys and Lophiocirsium).This work encompasses a remarkable number of species (n = 225), yet it is based only on two nuclear and five plastid markers that have proven to be insufficient for phylogenetic resolution in this genus [19].Moreover, according to the Supplementary Material [20], Table S2, the matrix lacks almost 50% of sequences (880 out of 1785).Moreover, the published phylogeny fails to support the classification proposed since Cirsium s. str is not recovered as monophyletic as the entire genus Carduus was nested within Cirsium s. str.In addition, two species of African Carduus, now Afrocarduus, are recovered within Cirsium s. str.[20], Figure S1, whereas our study reveals that African species of Carduus, now Afrocarduus, are a group evolutionary independent from Cirsium (Figures 1-3).
Interestingly, our plastid phylogeny agrees with the polyphyletic status of Cirsium, which was split into the same clades as in [20]'s phylogeny, and Carduus was nested within one of them.It is therefore likely that the results of [20] are strongly influenced by the plastid markers.In contrast, the classification in [20], Figure S1, is not supported by our nuclear phylogenies, where all species we included from sections Eriolepis (Cass.)Dumort.and Cirsium (according to [20]) are recovered as closely related and seem to share the same origin (Figures 1 and 2).Thus, our analyses based on nuclear data succeeded in recovering the natural groups within the group Carduus-Cirsium after segregating the African species, with high definition and strong support, probably thanks to our much larger molecular dataset obtained with the Hyb-Seq approach.In our opinion, this classification is the most conservative and morphologically consistent [14].Furthermore, given that Cirsium comprises more than 450 species with many practical applications, see [19], our conservative classification maintaining Cirsium as a single genus is also the most robust and operational one, because it avoids the inflation of hundreds of new nomenclatural combinations that would increase the already voluminous synonymy of Cirsium.
Additionally, hybrids between the two largest new genera proposed in [20] are extremely frequent [20,64].Hybridization between species and supra-specific entities constitutes an evolutionary driver and also a taxonomic and systematic challenge [65].This fact might also support that Cirsium should be treated as a single genus [19] until new studies covering the entire genus (diversity and distribution) are published.Meanwhile, the two subgenera, Cirsium and Lophiolepis Cass., can be distinguished by the absence or presence of setae on the upper leaf surface, respectively.A recent study (which partially covers the diversity and distribution) has provided some other characters that differ between the two subgenera: the monoploid genome size, genomic GC content and size of guard cells and achenes [64].
Finally, one species of Cirsium s. str.(Cirsium vulgare (Savi) Ten.) was recovered within sect.Eriolepis in our nuclear analyses (Figures 1 and 2).In our preliminary nuclear analyses, Cirsium vulgare had an unstable position between the two main clades of Cirsium (trees not shown), possibly because of a hybrid origin.This hypothesis is supported by the incongruence we detected between the plastome and nuclear phylogenies: in the plastome tree (Figure S2), Cirsium vulgare was recovered within Cirsium s. str., whereas it was recovered as a sister to a clade formed by all taxa of sect.Eriolepis in the nuclear trees (Figures 1 and 2).Similar results have been obtained for Cirsium italicum DC. [19], which could be of a hybrid origin [20,65].New analyses including a wider taxonomic sampling of the genus Cirsium are necessary to address a better infrageneric classification for Cirsium (Moreyra et al. in prep.).
The genus Carduus.The genus Carduus s. str.was recovered as a well-defined group in all our analyses (Figures 1 and 2 and Figure S2).Tyrimnus was recovered as a sister to Carduus, which is consistent with morphological characters.Originally, Tyrimnus was described as Carduus leucographus L. due to its morphological resemblance to Carduus.However, Tyrimnus can be well differentiated from Carduus from its s-shaped pappus bristles (straight in Carduus).We found the two species of Silybum to be a sister to Tyrimnus + Carduus.Within Carduus, we retrieved two clades that might deserve subgeneric status, but a final decision must await wider sampling and further analyses.

Figure 2 .
Figure 2. (A) Coalescence phylogenetic reconstruction for the subtribe Carduinae obtained with the nuclear dataset.Values above branches indicate Local Posterior Probability (LPP); only values above 0.95 are shown.Pie charts represent the proportion of the gene trees that support the main topology (blue), the first alternative (green) and the second alternative (red) for each node.(B) Coalescence phylogenetic reconstruction for the subtribe Carduinae obtained with the nuclear dataset.Values above branches indicate Local Posterior Probability (LPP); only values above 0.95 are shown.Pie charts represent the proportion of the gene trees that support the main topology (blue), the first alternative (green) and the second alternative (red) for each node.

Figure 3 .
Figure 3.Comparison of the phylogenies obtained with nuclear and plastid data.Clades are shown as collapsed at the generic level, following the classification proposed here.In the chloroplast concatenated tree, branches shortened for fit are indicated on the phylogeny as 1* (60% of reduction) and 2* (30% of reduction).The African species traditionally referred to as Cirsium dender I.Friis and C. englerianum O.Hoffm.formed a monophyletic group (Figure 1, BS = 100, TBE = 1; Figure 2, LPP = 1) outside the Carduus-Cirsium clade and as a sister to a clade formed by Galactites and Lamyropsis (Figure 1, BS = 100, TBE = 1; Figure 2, LPP = 1).The remaining tropical eastern African species traditionally referred to as Cirsium formed a monophyletic group (Figure 1, BS = 100, TBE = 1; Figure 2, LPP = 1) as a sister to the clade corresponding to subg.Afrocarduus (Figure 1, BS = 100, TBE = 1; Figure2, LPP = 1).In the concatenated phylogeny (Figure1), this clade (east African Cirsium + subg.Afrocarduus) was a sister to Picno-mon+Notobasis+non-African Cirsium, whereas in the coalescence phylogeny (Figure2), it was a sister to the main clade constituted by all the other lineages of the Carduus-Cirsium group.The genus Picnomon was recovered as a sister to Notobasis in the concatenated tree, but with low support (Figure1, BS = 65, TBE = 0.65), forming a lineage that was a sister to Cirsium s. str.In contrast, in the coalescence phylogeny, Picnomon was nested within the genus Cirsium with high support (Figure2, LPP = 1) and Notobasis was recovered with high support (Figure 2, LPP = 0.99) as a sister to the clade constituted by Silybum, Tyrimnus and Carduus s. str.The genus Tyrimnus (Figure 1, BS = 100, TBE = 1; Figure 2, LPP = 1) was recovered as a sister of Carduus s. str., whereas Silybum was recovered as a sister to the clade formed by Tyrimnus plus Carduus s. str. in both phylogenies (Figure 1, BS = 100, TBE = 1; Figure 2, LPP = 1).

Figure 3 .
Figure 3.Comparison of the phylogenies obtained with nuclear and plastid data.Clades are shown as collapsed at the generic level, following the classification proposed here.In the chloroplast concatenated tree, branches shortened for fit are indicated on the phylogeny as 1* (60% of reduction) and 2* (30% of reduction).

Table 1 .
Summary of the most accepted previous generic delimitation in the Carduus-Cirsium group and the new classification proposed here.