The genus Carthamus L. (Asteraceae, Carduoideae) consists of about 20 species with a distribution center in the Eastern Mediterranean and Irano-Turanian region1,2. Generic delimitation3,4 as well as intrageneric units are for a long time under debate. Hanelt5 divided Carthamus into five sections based on morphological characters, the same number was proposed by Estilai and Knowles6 based on cytological information. Vilatersana et al.3,7 using DNA marker regions divided the genus into two sections (section Carthamus and section Atractylis rchb.). This latter grouping was also followed by Bowles et al.1 and the split of Carthamus in two major clades was found multiple times with different molecular analysis methods [e.g., Refs.8,9] without formally referring to the sections. Section Carthamus species are assumed to be all diploid, while species within section Atractylis are polyploids. Unclear is the position of C. arborescens L.1 that sometimes is placed in the closely related genus Phonus hill3,4.

Safflower (Carthamus tinctorius L.), the only cultivated species of the genus, was domesticated in the Fertile Crescent over 4000 years ago10,11, most probably for its use as colorant. Nowadays it is mainly cultivated as an oilseed crop in suitable areas of Asia, northeastern Africa, Australia, the Americas and Europe2,12,13. With a total cultivation area of 1.1 million hectares a seed yield of approximately 1 million tons per year is reached14. Despite its growing commercial importance, genetic diversity and the parental origin of safflower are not fully clarified mainly due to limited taxon sampling among its close relatives included in phylogenetic studies, and partly also due to the low information content of fragment analyses for interspecific phylogenies such as RAPDs2,7, conserved intron-spanning PCR markers12, microsatellite markers15, ribosomal DNA RFLP analyses2, ISSR16, EST-SSR17 and combinations of these methods2. Some of the previous reports considered C. oxyacantha m.bieb., a species widespread in western Asia, as the wild progenitor of C. tinctorius17,18,19, while others found C. palaestinus eig from the Fertile Crescent the most likely progenitor12,20,21. The contribution of both C. palaestinus and C. oxyacantha in the origin of safflower has been reported based on chloroplast DNA diversity22. However, this latter study was based on limited taxon sampling, as just one individual from each of the wild species and two safflower individuals were included. In contrast to the phylogenetic analyses of diverse Carthamus species, within C. tinctorius analyses of length differences in PCR-amplified fragments mostly revealed high genetic diversity among safflower accessions2,7,12,16,23,24,25.

Here we use genome-wide single-nucleotide polymorphisms (SNP) obtained through genotyping-by-sequencing (GBS)26 on a diverse set of C. tinctorius accessions and six wild Carthamus species for a high-resolution phylogeny. The objectives of our study are (i) to infer phylogenetic relationships of Carthamus species closely related to safflower, i.e. section Carthamus, (ii) to confirm the closest wild relative(s) of safflower to enable breeding for the increase of genetic diversity regarding fatty acid content, drought tolerance and other agro-morphological traits, (iii) to look into genetic diversity of safflower accessions, and (iv) to see if the three to ten postulated morphology and geography-defined groups within C. tinctorius23,25,27,28,29,30,31 are genetically distinct and can be recovered using genomic data.


GBS sequencing

After removing the individual sample barcodes and quality filtering, our GBS data resulted in on average 1582 loci (min: 1409, max: 1614) per individual with 61 × coverage (min: 31 ×) for a dataset with 60 individuals covering safflower plus six wild Carthamus species with 3.5% missing sites.

For a larger dataset consisting of all C. tinctorius individuals together with three closely related wild species belonging to section Carthamus we obtained on average 7197 loci (min: 4087, max: 7513) per individual with a coverage of 163 × (min: 40 ×) with overall 8.7% missing sites.

Phylogenetic inference

In an initial phylogenetic analysis, we included six Carthamus species, i.e. C. boissieri halácy, C. glaucus m.bieb., C. lanatus L. (with three subspecies), C. oxyacantha, C. palaestinus and C. tenuis (boiss. & heldr.) hanelt together with a small but geographically diverse subset of C. tinctorius to infer the closest relative of safflower. The GBS-derived data matrix consisted of 108,985 aligned characters of which 3468 were parsimony informative. We used neighbor-joining (NJ), Bayesian phylogenetic inference (BI), maximum parsimony (MP) and SVDquartet (SVDQ) analyses which resulted in identical (BI, MP) or compatible (BI, MP, SVDQ, NJ) tree topologies. The MP analysis retained 2484 equally parsimonious trees with a consistency index (CI) of 0.68 and a retention index (RI) of 0.95. The strict consensus of these trees, summarizing MP (bootstrap, bs) and BI (posterior probabilities, pp) support values, is provided in Fig. 1, the SVDQ tree using the multi-species coalescent as tree model in Fig. 2.

Figure 1
figure 1

Unrooted strict consensus tree of 2484 MP trees derived from the analysis of GBS data. Asterisks indicate branches with support values of ≥ 98% in MP bootstrap analysis and posterior probabilities ≥ 0.99 in Bayesian phylogenetic inference for the backbone clades of the tree. The inserted tree on the left is one of the most parsimonious trees and indicates branch lengths in MP analysis.

Figure 2
figure 2

Unrooted SVDquartet species tree using the multi-species coalescent as tree model on the GBS dataset of the Carthamus target species. The individuals also analyzed in Fig. 1 were partitioned according to their species affiliation, with the introgressed C. × oxyacantha individuals treated as a separate category. Ploidy levels and chromosome numbers are listed behind the species names. Numbers along branches indicate bootstrap support values (≥ 50%).

These unrooted trees consist each of two major clades, one including the 24-chromosome species (chromosome numbers according to Vilatersana et al.3) C. oxyacantha, C. palaestinus, and C. tinctorius together with a paraphyletic 2n = 20 C. glaucus, while clade two harbors 2n = 20 C. tenuis and C. boissieri, together with the 2n = 44 allotetraploid C. lanatus. Carthamus tinctorius groups within a paraphyletic C. palaestinus, indicating the latter to be the closest relative of safflower and probably the wild progenitor of the crop. The two Iranian C. oxyacantha accessions from Isfahan and Kashan are separated from the other members of this species and fall within the C. palaestinus/C. tinctorius clade (Fig. 1), indicating introgression from these species. Support values for the branches separating the species are generally very high throughout the trees (Figs. 1, 2).

Since, C. tinctorius was found in a clade with C. oxyacantha, C. palaestinus and C. glaucus, we included these wild species in our analysis of 114 diverse safflower accessions, where we intended to analyze intraspecific groupings within C. tinctorius. The alignment for this dataset consisted of 435,024 characters with 7151 of them being parsimony informative. Maximum parsimony analysis of this dataset resulted in four MP trees (CI = 0.30, RI = 0.51) that were rooted with the six non-introgressed C. oxyacantha accessions. In this tree (Fig. 3) two clades of C. palaestinus group within C. tinctorius with 100% bootstrap support for the clade unifying these species. Within C. tinctorius bootstrap support values > 75% occur only for few and rather small clades of safflower accessions. The C. tinctorius accessions formed four larger groups (Fig. 3), which consisted however of three grades and only one clade. Thus, the borders of these groups and assignment of individuals to them is not unequivocal. We grouped safflower individuals initially into nine geographic regions according to their countries of origin and taking into account earlier proposed groupings5,10,27,29. No clear correlation between geographic patterns and phylogenetic groups was found within C. tinctorius, i.e. within most groups, safflower accessions from diverse geographical areas were united (Fig. 3). This was also the outcome of a SVDQ analysis for this dataset (not shown).

Figure 3
figure 3

One out of four equally parsimonious trees of a GBS-derived MP analysis of a large and geographically diverse set of 114 safflower accessions. The tree was rooted with C. oxyacantha. Bootstrap values ≥ 75% are indicated by asterisks at the branches. Gray branches depict clades that were not recovered in all MP trees and collapse in the strict consensus tree. Colored bars provide the assignment of safflower individuals to four intraspecific groups according to a population structure analysis. To the right of the groups the proportion of members derived from certain geographical areas are given.

Population assignment analysis and Principal Component Analysis (PCA)

In addition to tree-based estimations of relationships of C. tinctorius, we also used Bayesian population assignment and principal component analyses. Carthamus oxyacantha, C. palaestinus, C. glaucus and C. tinctorius individuals were subjected to structure-like analyses in lea. Evanno’s Δk suggested K = 3 as the best-fitting category number that was used for the final analysis (Fig. 4). The analysis confirmed our tree-based assumption that two of the C. oxyacantha accessions (from Isfahan and Kashan, Iran) were introgressed by genotypes from the C. palaestinus/C. tinctorius group and in addition identified also a third accession (from Arak, Iran) showing traces of gene flow. Carthamus palaestinus (from Israel) has the greatest similarity with C. tinctorius and its gene pool cannot be separated from C. tinctorius. Two individuals of C. palaestinus (No. 6 and 7) which grouped in tree-based analyses separate from the other five individuals of this species show slight traces of differing alleles (green) in this analysis (Fig. 4). Carthamus glaucus (from Lebanon) is clearly distinct from the former two species although some allelic overlap occurs with one Russian and two Iranian safflower accessions (Fig. 4) and two individuals of C. glaucus share alleles with C. palaestinus/C. tinctorius. PCA resulted in three groups of populations (Fig. 5). The first two principal components explained 42.8% of the total molecular variation. PCA results were similar to those of the phylogenetic trees and lea plots, as they grouped C. tinctorius and C. palaestinus together, while C. oxyacantha and C. glaucus were clearly separated from them along the first and second axis, respectively.

Figure 4
figure 4

Population structure analysis based on 3720 unlinked SNP loci in LEA with K = 3 for C. oxyacantha, C. palaestinus, C. tinctorius and C. glaucus.

Figure 5
figure 5

Principal component analysis (PCA) of C. oxyacantha (Co), C. glaucus (Cg), C. palaestinus (Cp) and C. tinctorius (Ct) along the first and second PCs.

To check the likely substructures within the C. tinctorius accessions and to gain a more precise understanding of the structural similarities between C. tinctorius and its close relative C. palaestinus, an additional set of population structure analysis was carried out including only C. tinctorius and C. palaestinus individuals. For K = 5, four major groups were revealed within C. tinctorius samples (Fig. 6). These four groups do not precisely reflect the four groups resulting from the tree-based analysis (Fig. 3) as they are partly mixed in these groups in the tree. However, also based on the population assignment analysis no clear geographic distribution pattern was observed for the major groups and admixture among the four groups is obvious. Most of C. palaestinus individuals occur together in a separate group, except for two samples where the majority of the allelic diversity is shared with C. tinctorius (Fig. 6).

Figure 6
figure 6

Population structure analysis based on 7556 unlinked SNPs in LEA with K = 5 for C. tinctorius and C. palaestinus individuals.

Analysis of genome size

Flow cytometry was used to measure genome sizes and infer ploidy levels of C. tinctorius, C. oxyacantha, and C. palaestinus individuals (Table 1). We found that all measured samples of C. oxyacantha, C. palaestinus and C. tinctorius represent diploid (2n = 2x = 24) 2C genome sizes ranging from 2.4 to 2.7 pg.

Table 1 Genome sizes of C. tinctorius, C. oxyacantha, and C. palaestinus individuals measured by flow cytometry against tomato cv. ‘Stupicke’ (2C = 1.96 pg) as size standard.


Inferring phylogenetic relationships among closely related species is a challenge because of inter-locus phylogenetic discordance and/or the difficulty of obtaining variable markers. Despite various phylogenetic studies carried out in Carthamus, phylogenetic relationships within Carthamus are not entirely clear and the closest wild relative of cultivated safflower is still debated due to analyses with limited taxon sampling and/or low information content of the applied markers2,4,8,22. Still, studies published during the last decades mostly identified C. palaestinus from the southwestern Fertile Crescent as the possible wild ancestor1,2,3, which fits geographically with the earliest archeological remains of safflower that were identified in Syria11.

Here we performed phylogenetic and population genomic analyses based on genome-wide distributed SNPs that were obtained through GBS to clarify the phylogenetic relationships of the relatives of C. tinctorius and to see if described ecotypes of safflower are genetically distinct. The GBS approach is known to have the potential for better-resolved relationships within and among closely related species in comparison to earlier fragment size analyses or sequence comparisons of single nuclear or chloroplast DNA regions32,33,34,35,36.

Our analyses of GBS data representing seven Carthamus species resulted in phylogenetic trees (Figs. 1, 2) where C. boissieri, C. lanatus and C. tenuis form one clade that is sister to the clade of C. glaucus, C. oxyacantha, C. palaestinus and C. tinctorius. Support values for the clades in the trees are generally very high and the individuals of C. boissieri, C. lanatus, most C. oxyacantha, and C. tenuis all form monophyletic groups (Fig. 1). In contrast, C. tinctorius individuals group together within C. palaestinus, and both together within a paraphyletic C. glaucus. Within the C. palaestinus/C. tinctorius clade also two C. oxyacantha accessions grouped, which we assumed to be hybrids between C. oxyacantha and C. palaestinus or C. tinctorius, a result that was confirmed by structure-like analysis. Also in C. glaucus two individuals show allelic similarities with C. tinctorius, which could indicate gene flow and explains that two clades of a paraphyletic C. glaucus were recovered in the phylogenetic trees of which the introgressed individuals are sisters to C. palaestinus/C. tinctorius. This introgression signal could also explain the reduced support values in the SVDQ tree for the respective clades (Fig. 2). In C. palaestinus two individuals group apart from the other members of this species (Figs. 1, 3) and show differences in their allelic patterns (Figs. 4, 6) although they belong to the same accession. It is unclear if this reflects the original diversity of the source population or introgression during ex situ conservation in a gene bank. Our data indicate that hybridization between the species closely related with C. tinctorius is still possible on the homoploid level, as the C. oxyacantha individual from Isfahan (Iran) was measured with the same genome size as other individuals from this species. The two major clades we obtained in our phylogenetic tree essentially reflect the two sections Carthamus and Atractylis of Bowles et al.1. However, in our analysis C. glaucus groups with the section Carthamus species while in Bowles et al.1 as well as in Vilatersana et al.3 and Tarıkahya-Hacıoglu et al.37 C. glaucus is a member of the sect. Atractylis taxa. This inconsistency is however not unique to our analysis. Already in Sasanuma et al.8, C. glaucus is grouping with C. tinctorius while Sehgal et al.2 and Mehrotra et al.9 retrieved C. glaucus accessions in both clades of Carthamus. Unfortunately, we relied on gene bank passport data and had no additional seeds available when we obtained the GBS results so we could not check chromosome numbers for our C. glaucus individuals to be sure that they are indeed belonging to the 2n = 20 chromosomes taxa. As in all before-mentioned studies, C. glaucus materials from gene banks were used instead of freshly collected plants from the wild. It might therefore be worthwhile to re-determine gene bank-derived C. glaucus accessions in future analyses to be sure about the species affiliation of these lineages.

Intra-species relationships within safflower are not well understood. Based on morphological traits like flower color, inflorescence size, plant height and leaf shape, and their geographical distribution, between five and ten centers of diversity were postulated for safflower, all harboring region-specific ecotypes1,9,18,19,25. In our second GBS dataset we included a diverse set of 114 C. tinctorius accessions together with the section Carthamus species of our initial analysis but excluded the two obviously introgressed C. oxyacantha accessions. While C. glaucus appears again paraphyletic in the resulting phylogenetic tree (Fig. 3), the individuals of C. palaestinus fall within the safflower accessions occurring in two of their different clades. As the C. palaestinus individuals in our study show the darker red to brownish flower color (in contrast to the orange flowers of safflower) and seeds are small compared to the crop, we assume that the plants are from a true wild population and do not belong to feral individuals escaping cultivation. This result shows that the gene pool of C. tinctorius cannot be separated from that of C. palaestinus and is a strong indication that the latter is the wild progenitor of safflower, as its gene pool overlaps with C. tinctorius. This result is supported by the population assignment analysis where C. palaestinus has the highest similarity with C. tinctorius (Fig. 6), and confirms earlier findings regarding the ancestor of safflower2,12,20.

Within safflower we found no clear subgroups. Carthamus tinctorius in the tree provided in Fig. 3 has four groups of genetically similar accessions. Four groups were also obtained in population assignment analysis (Fig. 6). However, only one of the groups is monophyletic in the phylogenetic tree and none of them shows a clear geographic pattern and none correlates strictly with the groups obtained in the LEA analysis (Fig. 3). Expectedly, bootstrap values within this phylogenetic tree are low, often not even reaching 50%. We interpret this result as an indication of strong trading connections among the early centers of safflower cultivation in Eurasia, followed by multiple independent introductions of safflower germplasm from these centers into regions like the Americas and Australia where safflower cultivation started only during the last century. In addition to the mere introduction of foreign germplasm to new areas also gene flow between newly attained and older local cultivars then further blurred phylogeographic patterns. Admixture among the four different genetic groups of safflower is obvious in most individuals in the population assignment analysis (Fig. 6). This general picture does not change when C. palaestinus, C. oxyacantha and C. glaucus individuals with traces of safflower introgression are removed from the dataset: in this case the resulting clades and subclades in the phylogenetic analysis again lack geographic structure (see Supplementary Fig. S1). This result is in strong contrast to the morphology-defined geographic groups of Ashri29, Hanelt5, Knowles and Ashri10, and Kupcov27, which all postulated the presence of certain types typical for specific geographic regions, although they did not arrive at a consensus about the number of such safflower types. This discrepancy can be explained either by the quantitative nature of the used morphological traits that are (i) controlled by very few genetic loci with the respective allelic composition not reflected in the GBS data we obtained and/or (ii) these traits might have an environmental and ecological component that influences characters like flower color or plant height, creating ecotypes where genetic inheritance for the traits might be low. Still, in contrast to our results here Chapman et al.20, using 24 nuclear microsatellite markers, found five very weakly supported groups with a possible geographic correlation. In contrast to our dataset, they obtained these groups by restricting their analysis to Ashri’s29 set of assumed autochthonous accessions, i.e. a set of materials that were not influenced by introductions during the last century.

For the safflower accessions currently stored in gene banks, we conclude that neither geographic origin nor certain morphological traits are usable indicators of genetic similarity or diversity. For breeding approaches, it is therefore necessary to either use molecular analyses to identify rather diverse genotypes or use gene bank passport data to identify accessions with the preferred or extreme agronomic traits. Moreover, the collections stored at the German National (IPK; identified by “CART” in their accession numbers) and the US Department of Agriculture (USDA; identified by accession numbers starting with “PI”) gene banks seem only marginally overlapping, as two of the phylogenetic groups consist mainly of USDA and two of IPK accessions (Fig. 3). We screened, however, only a part of these collections so that this can also be an artifact of our accession selection.


Using GBS that combines next-generation sequencing with a genomic complexity reduction approach, phylogenetic relationships of all analyzed safflower accessions could be resolved. However, within C. tinctorius we could not recover any geographically defined groups/ecotypes that were postulated earlier based on morphological character combinations thought to prevail in specific regions. This we explain with seed exchange among the areas where safflower is commercially grown, which might have blurred an eventually formerly existing geographic pattern of locally evolved genotypes.

In contrast to the intra-species structure in C. tinctorius our GBS analysis recovered species relationships in C. sects. Carthamus and Atractylis with high support values. We identified C. palaestinus as closest relative and likely progenitor of safflower. Carthamus glaucus resulted as a paraphyletic species that harbors C. tinctorius/C. palaestinus, but its phylogenetic position in our trees casts some doubts about the species determination of the used material. The other analyzed species appear to be monophyletic.

Materials and methods

Taxon sampling

One hundred sixty-seven individuals (see Supplementary Table S1) including 114 cultivated C. tinctorius and 53 individuals of wild species of Carthamus, i.e. C. boissieri (8), C. glaucus (6), C. lanatus (18), C. oxyacantha (8), C. palaestinus (7) and C. tenuis (6), were included in GBS-based phylogenetic analyses. A set of eight diverse individuals of C. tinctorius was used together with individuals of C. palaestinus, C. oxyacantha and C. glaucus for population structure analyses. The analyzed materials were obtained as seeds from the collections of RTIPP-SBUK (Research and Technology Institute of Plant Production, Shahid-Bahonar University, Kerman, Iran), IPK (Leibniz Institute of Plant Genetics and Crop Plant Research, Gatersleben, Germany) and NPGS (National Plant Germplasm System, U.S. Department of Agriculture, USA) and grown at least to a four-leaf stage before harvesting leaves for DNA extraction. We assigned the safflower accessions to nine geographic regions according to their country of origin following roughly the earlier postulated groupings5,20,29. These are Southwest Asia (SW Asia: Levant, Turkey, Iraq), Central Asia (C Asia: Kazakhstan, Tajikistan, Uzbekistan), East Asia (E Asia: China, Korea and Japan), Southeast Asia (SE Asia: India and neighboring countries), Europe (including accessions from Russia), Iran (Iran, Afghanistan), Africa, the Americas and Australia. Detailed information such as passport data and observation records for each accession can be found in the online GRIN (Germplasm Resources Information Network) database of NPGS ( and Genebank Information System of the IPK Gatersleben ( Collection and use of plant materials was conducted in accordance with institutional, national and international regulations. Seed exchange followed the Nagoya rules of access and benefit sharing through standard material transfer agreements.

DNA extraction and genotyping-by-sequencing (GBS)

Genomic DNA was extracted from young leaves with a DNeasy Plant Mini kit (QIAGEN). To obtain genome-wide SNPs, we used a two-enzyme GBS protocol that uses the restriction enzymes PstI (NEB Inc.) and MspI (NEB Inc.) to digest 200 ng of genomic DNA. For the library preparation and individual barcoding, the protocol of Wendler et al.38 was followed. The individual libraries were sequenced in multiplex on an Illumina HiSeq 2500 (100 bp single-end reads) at the sequencing facilities of Leibniz Institute of Plant Genetics and Crop Plant Research (IPK), Germany targeting a minimum coverage of at least 40×.

GBS data preparation and filtering

Initial quality assessment of all raw sequence samples was performed using fastqc39 to check for overrepresented reads, remaining adapter sequences and to decide on trimming thresholds. Sequences were de-multiplexed allowing for one base mismatch in barcodes, and the restriction site and barcode were trimmed from all GBS sequence reads using cutadapt40. Sequence data were quality filtered and clustered using the ipyrad bioinformatics pipeline41 with a threshold value for coverage of at least 30×. We conducted two runs of ipyrad resulting in two GBS alignments concatenating all filtered loci: (i) one with a small set of C. tinctorius accession together with six wild Carthamus species, and (ii) using 114 individuals from C. tinctorius plus its closest three relatives according to the prior analysis. The minimal number of samples to possess a certain locus was set to 90% of the individuals. The clustering threshold of reads within and between individuals was set to 0.85 and 0.90 in the first and second datasets, respectively. For the other parameters, the default settings were used. The assembly of the GBS data was done de novo. SNPs were checked for their patterns and found to be nearly completely bi-allelic, even in tetraploid C. lanatus. SNP calling could therefore be conducted in ipyrad that has only a setting for diploid taxa.

Phylogenetic analyses

We used mrbayes 3.2.642 and paup* 4a16943 to infer phylogenetic relationships within our set of Carthamus species and accessions. NJ, based on General Time Reversible (GTR) distances, MP and SVDquartet trees were calculated in paup* to analyze the aligned sequence data matrices. In MP gaps were treated as missing data and we used a two-step heuristic search with Tree Bisection/Reconnection (TBR) branch swapping as described in Blattner44 with initial 500 random addition sequences (RAS) restricting the search to 50 trees per replicate. The resulting trees were afterwards used as starting trees in a search with maxtree set to 10,000. To test clade support, bootstrap analyses were run with re-sampling the dataset 500 times with the same settings as before, except that we did not use the initial RAS step. For both datasets SVDquartets were calculated in paup* running 500 bootstrap re-samples. Individuals were partitioned according to their species affiliation in the smaller dataset including all species while for the dataset with all C. tinctorius individuals no partitioning was used. In these analyses trees were selected using Quartet Fiduccia and Mattheyses (QFM) assembly and the multi-species coalescent (MSC) as the tree model. paup* was also used to infer the model of sequence evolution using the Bayesian information criterion.

For BI two times four chains were run for 5 million generations for the dataset including the set of different Carthamus species, specifying the model of sequence evolution as GTR + Γ. We sampled a tree every 1000 generations and summarized the trees in mrbayes. Converging log-likelihoods, potential scale reduction factors for each parameter and inspection of tabulated model parameters in mrbayes suggested that stationary had been reached. The first 25% of trees were discarded as burn-in. For the large dataset including all C. tinctorius accessions, phylogenetic inference calculations in mrbayes were run with different MCMC settings each for more than three weeks (about 10 million generations) without converging chains when we cancelled these analyses.

Population structure analyses

The dataset of unlinked SNPs generated from ipyrad was used as input for a Principal Component Analysis (PCA) within ipyrad and for the analysis of population structure for C. oxyacantha, C. palaestinus, C. glaucus and eight C. tinctorius cultivars in the R package LEA45. The number of subpopulations (K) was tested for a range from one to seven, resulting in an optimal K = 3 for this dataset. An admixture analysis46 using sparse nonnegative matrix factorization (snmf) was used for the estimation of population genetic structure. The same analysis approach was used for a dataset including all C. tinctorius accessions together with the individuals of C. palaestinus. In this case, we arrived at an optimal K = 5.

Analysis of genome size

To obtain ploidy levels we used flow cytometry to measure the genome sizes of Carthamus species47. For this, we collected a leaf from each of three individuals from eight C. tinctorius and four C. oxyacantha accessions and three leaves from one C. palaestinus individual. Afterwards, the leaves were transported to the lab and genome sizes were measured on a CyFlow (Sysmex Partec) flow cytometer using propidium iodide (PI) as the staining reagent (Sigma-Aldrich) according to Jakob et al.48. The tomato cultivar ‘Stupicke’ was used as the size standard (2C DNA content = 1.96 pg49).