DNA sequence diversity and the origin of cultivated safflower (Carthamus tinctorius L.; Asteraceae)

Background Safflower (Carthamus tinctorius L.) is a diploid oilseed crop whose origin is largely unknown. Safflower is widely believed to have been domesticated over 4,000 years ago somewhere in the Fertile Crescent. Previous hypotheses regarding the origin of safflower have focused primarily on two other species from sect. Carthamus – C. oxyacanthus and C. palaestinus – as the most likely progenitors, although some attention has been paid to a third species (C. persicus) as a possible candidate. Here, we describe the results of a phylogenetic analysis of the entire section using data from seven nuclear genes. Results Single gene phylogenetic analyses indicated some reticulation or incomplete lineage sorting. However, the analysis of the combined dataset revealed a close relationship between safflower and C. palaestinus. In contrast, C. oxyacanthus and C. persicus appear to be more distantly related to safflower. Conclusion Based on our results, we conclude that safflower is most likely derived from the wild species Carthamus palaestinus. As expected, safflower exhibits somewhat reduced nucleotide diversity as compared to its progenitor, consistent with the occurrence of a population genetic bottleneck during domestication. The results of this research set the stage for an investigation of the genetics of safflower domestication.


Background
Safflower (Carthamus tinctorius L.) is a thistle-like, selfcompatible, annual, diploid (2n = 24) herbaceous crop that thrives in hot, dry climates, and is capable of surviving on minimal surface moisture. It is believed to have been domesticated somewhere in the Fertile Crescent region over 4,000 years ago [1]. Following its initial domestication, safflower cultivation is thought to have expanded to both the east and west [2], with Knowles [3] ultimately recognizing seven "centers of similarity" (the Far East, India-Pakistan, the Middle East, Egypt, Sudan, Ethiopia and Europe). Safflower lines native to each 'center' are remarkably similar in height, branching, spines, flower color and head size; however, consistent morphological differences are maintained between the centers.
For centuries, safflower was grown on a local scale for its flowers, which served as a source of dye (carthamine) for textiles and food coloring, as well as for use in religious ceremonies [4]. Floral extracts were also used to flavor foods, and have historically been valued for their numerous medicinal properties. Cultivation of safflower in the New World commenced in 1899, and commercial pro-duction of safflower as an oilseed crop began in the 1950s [5]. More recently, there has been growing interest in safflower for its potential as a large-scale production platform for plant-made pharmaceuticals [6,7].
To date, phylogenetic investigations of Carthamus species have focused on either delimiting the sections within the genus [e.g. [8,9]], or on the development of DNA fingerprinting methodologies for investigating relationships amongst safflower cultivars [e.g. [10]]. Relationships between closely-related species within each section are, however, only poorly understood, and details surrounding the origin and early evolution of safflower are lacking. What we currently know is that safflower belongs to a group of closely related diploid species (sect. Carthamus; all 2n = 24 chromosomes [11]) whose ranges extend from central Turkey, Lebanon, and Israel in the west to northwestern India in the east. In addition to C. tinctorius, this section is composed of C. curdicus Hanelt, C. gypsicola Iljin, C. oxyacanthus Bieb. (= C. oxyacantha M. Bieb.), C. palaestinus Eig, and C. persicus Desf. ex Willd. (= C. flavescens Spreng) [8,12]. However, these species all exhibit some degree of cross-compatibility with one another [reviewed in [12,13]] and thus reproductive isolation alone cannot be used to delimit the species. Carthamus curdicus and C. palaestinus exhibit restricted geographical distributions (Northern Iran and Southern Israel, respec-tively), whereas C. persicus, C. gypsicola and C. oxyacanthus are more widely distributed throughout the Middle East [12].
Hypotheses regarding the origin of safflower have focused on C. oxyacanthus or C. palaestinus as the most likely progenitors, although C. persicus has also been suggested as a possible progenitor [14]. Here we report on levels of nucleotide diversity within and among species of sect. Carthamus, and investigate the origin of cultivated safflower using data derived from seven nuclear genes.

DNA sequence diversity
Sequence data were collected from all seven gene regions for each of the 23 individuals surveyed (Tables 1 and 2), encompassing all species within sect. Carthamus. All sequences have been deposited in the Genbank Data Library and are available under accession nos. EF483951-EF483974, EF483983-EF484014, EF519712-EF519729, EF519732-EF519751, EF519754-EF519770, EF519774-EF519792, EF519795-EF519811, EF519815-EF519834 and EF519838-EF519857. Excluding indels, sequence lengths varied from 365 to 621 base pairs (bp) per locus, and all sequences included both exons and introns (Table  3). Thus, we were able to analyze 3239 bp of aligned sequence per individual with 1317 bp (40%) coming from exons and 1922 bp (60%) coming from introns. Across taxa, the number of indel polymorphisms per locus varied from four to thirteen, with a total of 53 indels in the data set. All indels were excluded from the analyses of nucleotide polymorphism.
Single nucleotide polymorphisms (SNPs) were considerably more common than indels. A total of 220 SNPs were found across the full data set, resulting in an average of 1 SNP per 15 bp of sequence. Considering just the 11 safflower individuals, there were 34 SNPs, corresponding to an average of 1 SNP per 95 bp of sequence. Estimates of nucleotide diversity for C. tinctorius, C. palaestinus and C. oxyacanthus are presented in Table 4.
Because of the relatively small amount of exonic sequence included in each gene fragment (188 bp on average), individual estimates of synonymous and non-synonymous diversity must be viewed with caution. Averaging across loci, however, revealed that our estimates of non-synonymous variability are substantially lower than our estimates of synonymous variability, suggesting that diversity at these loci is primarily governed by purifying selection (data not shown). After correcting for multiple comparisons, none of the Tajima's D estimates were significantly different from zero (Table 4). Primer names followed by 'a' or 'b' were designed to amplify a given locus in two parts. Functional annotations were taken from the Genbank record for each Arabidopsis locus. For the unknown protein a BLASTn search failed to yield a putative function

Phylogenetic relationships
Comparison of the NJ trees produced from the single gene analyses suggests that reticulate evolution and/or incomplete lineage sorting has occurred in Carthamus sect. Carthamus ( Fig. 1). While a number of overall similarities in tree topology are evident from these analyses, there are several instances in which the phylogenetic position of an individual varies depending on the gene analyzed. In some cases, individuals harbored divergent alleles at one or more loci, possibly indicating that contemporary hybridization has played an active role in the evolution of sect. Carthamus. Of particular note are the positions of the C. gypsicola alleles which are sometimes found in diver-gent clades (e.g., for genes A19 and B12). Some C. oxyacanthus alleles show a similar pattern (e.g., genes A39 and B27). When comparing the seven trees for the individual loci, however, some patterns begin to emerge. Overall, C. oxyacanthus is most often found to be relatively distantly related to C. tinctorius, and frequently associated with alleles from C. gypsicola. Of particular note is the close relationship between individuals of C. tinctorius and C. palaestinus, suggesting that the species most closely related to safflower (and hence its most likely progenitor) is C. palaestinus (Fig. 1).  Despite the occurrence of some incongruities between loci, the NJ, ML and Bayesian trees based on the combined data (which are nearly identical to each other in topology; Fig. 2 and data not shown) are in overall agreement with our interpretations of the single gene analyses. Carthamus oxyacanthus is resolved as the species most distantly related to safflower, with high ML bootstrap and Bayesian posterior probabilities. Within C. oxyacanthus the two lines from Pakistan are more closely related to each other than they are to the line from Afghanistan. Carthamus persicus appears to be paraphyletic, perhaps due to recent gene flow. As predicted from the individual gene trees, C. palaestinus is the most closely related species to safflower, and we conclude that this species is the most likely progenitor of safflower. In the Bayesian analysis, all four C. palaestinus individuals are found in a well-supported clade along with all of the safflower individuals. Similarly, in the ML analysis, three of the four individuals of C. palaestinus are found in such a clade (87% BS), with the fourth resolving at the base of this clade along with individuals of C. curdicus and C. persicus. Some relationships can also be resolved among safflower individuals; for example, saf-fAZ and saffAC form a well-supported clade at the base of the safflower/C. palaestinus group. Relationships between the other cultivars are, however, poorly-supported.

Origin of safflower
A close relationship between members of sect. Carthamus has been proposed based on data from crossing studies [reviewed in [12,13]], the identification of natural hybrids amongst some species within the section [1,14], and phy-Phylogenetic relationships among species of Carthamus sect. Carthamus based on single-gene analyses Figure 1 Phylogenetic relationships among species of Carthamus sect. Carthamus based on single-gene analyses. Neighbor-Joining trees were generated for each individual gene. Species names and accession codes are given in Table 1. Accession names followed by -1 or -2 denote alleles for a given locus. Alleles followed by a * were determined using haplotype subtraction by maximum likelihood; the remainder of the alleles were determined by cloning. logenetic analyses involving some members of the section [9]. Prior to the present investigation, however, the phylogenetic relationships amongst all species within sect. Carthamus had not been investigated, and the identity of the progenitor of safflower had only been hypothesized. While Ashri & Knowles [14] proposed that safflower was derived from hybridization between C. oxyacanthus and C. persicus, this hypothesis is clearly not supported by our results. Rather, Carthamus palaestinus and safflower are found in the same clade indicating a close relationship between these species. We thus propose that C. palaestinus, which is native to the deserts of southern Israel and western Iraq, is the wild progenitor of safflower. Safflower and C. palaestinus share a self-compatible breeding system [12]; thus, the near absence of heterozygous loci in these species (Fig. 1) is unsurprising. Carthamus persicus and some populations of C. oxyacanthus are self-incompatible, and this is evident from the presence of much more heterozygosity in these species (Fig. 1). The cause of the nonmonophyly of C. persicus remains unknown due to the small sample sizes necessarily employed here; however the retention of ancestral polymorphism and/or contemporary gene flow (C. persicus is self-incompatible) could be responsible.
As noted in the Introduction, Knowles [3] recognized seven distinct "centers of similarity" of safflower, including the Far East, India-Pakistan, the Middle East, Egypt, Sudan, Ethiopia and Europe. Interestingly, our data provide little support for the distinctiveness of safflower accessions from these disparate geographic locales. Indeed, while there is a small amount of phylogenetic Phylogenetic relationships among species of Carthamus sect. Carthamus based on a combined analysis of seven nuclear genes  Table 1. B. Bayesian structure apparent within safflower, it appears that most of the accessions surveyed herein are highly similar at a genetic level. Moreover, much of the substructuring within safflower is not well-supported in either the Bayesian or ML analyses. The exceptions to this are a pair of accessions from the USA and Canada (saffAZ and saffAC, respectively), and possibly one accession from Egypt (saff1067). Considering what we know about the history of safflower cultivation, the North American accessions (saffAZ, saffAC, and saffU) are presumably recent introductions, and from our data it appears that they may be derived from relatively divergent ancestral stocks (Fig. 2). Further investigation of the 'seven centers' hypothesis, will require the development and application of more variable markers to a much more robust sampling of the available safflower germplasm. While a recent investigation of safflower cultivars using RAPDs, ISSRs and AFLPs revealed some genetic structuring within the species [10], the authors did not address the question of whether or not the seven morphological centers of diversity correspond to genetic subgroups within safflower.

Levels of nucleotide diversity
The domestication of plant species is typically accompanied by a reduction in genetic diversity resulting from the population genetic bottleneck that occurs during domestication. Although results vary across species, crops generally harbor ca. two-thirds of the diversity that is present in their wild progenitors [15]. We found a similar value here, with a 20-30% (depending on the measure) reduction in nucleotide diversity in safflower as compared to C. palaestinus (Table 4). Because θ W is roughly proportional to heterozygosity, we can further conclude that a randomly selected pair of safflower sequences will differ at an average of 1 out of every 303 bp (i.e., 1/0.0033 ≅ 303). This makes safflower considerably less diverse than crops such as maize (1 out of every 105 bp [16]) and sunflower (1 out of every 140 bp [17]) but far more diverse than crops such as soybean (1 out of every 1030 bp [18]).

Conclusion
Insights into the origin of crop plants and knowledge of the identities of their progenitors are of great value in both basic and applied research programs. For example, the comparative analysis of crop plants and their wild progenitors can shed light on the genetic mechanisms underlying organismal evolution [19,20]. Similarly, comparative analyses of this sort can be a powerful tool for identifying genes underlying agronomically-important traits [21][22][23].
The identification of C. palaestinus as the wild progenitor of safflower opens the door for such analyses within the genus Carthamus. Moreover, because safflower is a modern-day oilseed crop and a member of the same family as cultivated sunflower, which has been the subject of a great deal of recent study [24,25], the initiation of such work in safflower would create a comparative framework for studying the evolution of oilseed crops within the Asteraceae.

Plant materials and DNA extraction
Tissue for DNA extraction was either obtained from live plants grown from seed or from herbarium specimens ( Table 1). Seeds were obtained from archived collections held at the USDA Western Regional Plant Introduction Station in Pullman, WA. These included 11 accessions of C. tinctorius L., three accessions of C. oxyacanthus and one accession of C. palaestinus. In addition Dr. R. Vilatersana (Institut Botànic de Barcelona) kindly provided seeds of C. gypsicola and C. persicus. Herbarium specimens of C. palaestinus (three accessions) and C. persicus (one accession) were provided by the IPK Gatersleben Herbarium (GAT), and C. curdicus (one accession) was provided by the Vienna Museum of Natural History (Herbarium W). All species are presumed to be diploid based on prior investigations [reviewed in [11]].
For the live plants, seeds were clipped with a razor blade and germinated on damp filter paper in Petri dishes (48 hours dark/48 hours light). Seedlings were then planted in soil and grown in the greenhouse. Total genomic DNA was then isolated from 100 mg of leaf tissue using the DNeasy plant mini kit (Qiagen, Valencia, CA). For the herbarium extractions, tissue lysis and extraction followed the same protocol as the fresh leaf tissue except that only ca. 20 mg of leaf tissue was used.

Locus selection and sequencing
The seven nuclear genes used in this study (Table 2) were selected from a set of universal markers that were recently developed for use in the Asteraceae [26]. The loci selected for inclusion in this study all produced a single amplicon that could be sequenced directly (only those individuals that were heterozygous for insertions/deletions were cloned). Three of the loci (A25, A39, and B27; Table 2) did not amplify well in the herbarium material, presumably due to DNA degradation; internal primers were thus designed to amplify these loci in two overlapping segments that were later aligned into a single contig. For A39 the first portion still could not be amplified in the herbarium specimens, and hence was scored as missing data for those individuals.
PCR was performed in a 20 µl total volume containing 20 ng of template DNA, 30  For individuals that were heterozygous for indels at a particular locus (as evidenced by the initial sequencing chromatogram), the unpurified PCR product was cloned using the pDrive (Qiagen) or TOPO TA (Invitrogen, Carlsbad, CA) cloning vectors following the manufacturers' protocols. In order to protect against Taq errors, PCR products from five positive clones per cloning reaction were then prepared and sequenced as above, except that the T7 and M13 universal vector primers were used.

Data analyses
DNA sequences were edited using Chromas 2.12 (Technelysium, Helensvale, Australia). Heterozygous bases from uncloned PCR products were detected by the presence of double peaks and coded following the conventions of the International Union of Biochemistry and Molecular Biology. From these, haplotypes were resolved using the maximum likelihood algorithm PL-EM [27] within the HapAnalyzer software [28]. Sequences were aligned using Clustal W2 [29] with the default settings, followed by manual adjustments. Indels were scored as additional characters using GapCoder [30], although regions that could not be aligned unambiguously and length variants at simple-sequence repeats were excluded from the analysis. Heterozygotes were common, and alleles were kept separate for the individual gene phylogenetic analyses. For the combined dataset, however, the phase of alleles across loci could not be reliably determined. As such, pairs of cloned alleles were collapsed into a single genotype for each gene and then the seven genes were concatenated for each individual.
Estimates of nucleotide diversity (π and θ, calculated on a per-site basis) as well as Tajima's D [31] were obtained for the three taxa with three or more samples (C. tinctorius, C. palaestinus, and C. oxyacanthus) using the software package DnaSP 4.00.5 [32,33]. Neighbor-Joining trees were generated separately for each locus and for the combined dataset using PAUP* ver. 4.0b [34]. The combined dataset was also subjected to Maximum likelihood (ML) and Bayesian analyses. ML analysis was carried out using PHYML v2.4.4 [35] under the HKY+Γ model of molecular evolution with four substitution rate classes with 500 bootstrap replicates. Bayesian analysis was conducted with MrBayes [36] as implemented in the Geneious package (v3.0.6; Biomatters Ltd., Auckland, New Zealand). MCMC analysis was run with four chains simultaneously for 1,100,000 generations, subsampling every 200 generations. Samples prior to the generation 100,000 were treated as burn-in and discarded.

Authors' contributions
MAC and JMB conceived the investigation, carried out the analyses and wrote the paper. MAC carried out the PCR and sequencing. All authors read and approved the final manuscript.