Background

Saussurea DC. is one of the largest genera in the family Asteraceae [1, 2]. It comprises approximately 300 species that are distributed in Asia, Europe and North America, with the highest diversity in the Himalayas and central Asia [3, 4]. Saussurea exhibits extreme morphological diversity and exists in habitats ranging from steppes to moist forests to cold and dry alpine meadows above 5000 m [3, 5].

Several phylogenetic studies have been conducted on Saussurea but the circumscription and infrageneric relationships of the genus remain controversial [5,6,7,8,9,10,11,12]. Lipschitz [13] recognized a total of 390 species belonging to six subgenera, namely subg. Saussurea DC., Jurinocera (Baill.) Lipsch., Eriocoryne (DC.) Hook. f., Amphilaena (Stschegl.) Lipsch., Theodorea (Cass.) Lipsch. and Frolovia (DC.) Lipsch. However, molecular evidence [7, 10, 14] has indicated that subg. Jurinocera, subg. Frolovia and sect. Elatae of subg. Saussurea should be excluded from Saussurea and treated as independent genera: Lipschitziella R.V. Kamelin, Frolovia (DC.) Lipsch. and Himalaiella Raab-Straube, respectively. Using molecular and morphological evidence, Shi and Raab-Straube [3] suggested Saussurea sect. Aucklandia should be treated as a new genus, Aucklandia Falc. Based on sequences of five loci (rbcL, ndhF, matK, trnL-F and ITS) and morphology, Wang et al. [11] established the genus Shangwua Y. J Wang, Raab-Straube, Susanna & J. Q. Liu from sect. Jacea, leaving four subgenera (Saussurea, Eriocoryne, Amphilaena and Theodorea) as constituting Saussurea s.s. Despite this progress, the relationships among and within these four subgenera have been poorly resolved due to a potentially rapid radiation, leaving insufficient phylogenetic signal at deeper nodes [5]. No phylogenomic studies have yet assessed these relationships, although a recent study using target enrichment of nuclear genes to resolve Cardueae relationships sampled 19 species of Saussurea representing two subgenera [15].

Plastomes have been proven to be powerful tools for exploring deep relationships in the plant Tree of Life [16,17,18,19]. They have helped resolve ambiguous relationships of particularly recalcitrant lineages, such as those that have undergone rapid evolutionary radiations (e.g. [20,21,22,23]). Complete plastome sequences also provide insight into the molecular evolutionary patterns associated with gene rearrangements, duplication and loss (e.g. [24,25,26]), and in some cases these structural changes are phylogenetically informative characters in and of themselves, as for example the two large inversions (~ 20 kb and ~ 3 kb inversions) that characterize the Large Single-Copy (LSC) region of most Asteraceae plastomes [27,28,29].

Different regions of the plastome have different selective constraints that may yield differing estimates of phylogeny, as for example noncoding versus coding regions [30, 31]. Selective forces may also play a role in driving plastome structure [32], including rearrangements [33] and gene loss [34, 35]. However, the effect of selective forces in plastome evolution within Asteraceae remains unclear.

To date, only three Saussurea plastomes have been reported: S. involucrata [36], S. chabyoungsanica [37] and S. polylepis [38]. Here, we sequenced 17 species representing all four subgenera of Saussurea in order to (1) elucidate plastome evolution, including structural variation and molecular signals of selection, (2) estimate the effectiveness of different plastome data sets in resolving relationships within this radiating lineage, and (3) investigate the backbone relationships within Saussurea.

Results

Characteristics of Saussurea plastomes

After de novo and reference-guided assembly, we obtained a single scaffold for each plastome. The sequencing and assembly information are provided in Tables 1 and Additional file 1: Table S2. The sizes of the 17 Saussurea plastomes were similar, ranging from 151,474 bp in S. tridactyla to 152,658 bp in S. przewalskii. All 17 plastomes possessed the typical angiosperm quadripartite structure and contained 113 unique genes, including 79 protein-coding genes, 30 transfer RNA (tRNA) genes and four ribosomal RNA (rRNA) gene. A total of 18 genes (including 11 protein-coding genes and 7 tRNA genes) had introns, with 15 genes having one intron and three genes having two introns. The IR regions were also highly consistent, all of which included 17 genes (six protein-coding genes, seven tRNA genes, and four rRNA genes). In all plastomes, the rps12 gene was found to be trans-spliced, with one of its exons located in the LSC region and the other duplicated in the IR (Fig. 1).

Table 1 Characteristics of newly sequenced plastomes
Full size table
Fig. 1
figure 1

General plastome map of Saussurea. Specific sizes for the plastomes of each species are presented in Table 1. Inv: inversions

Full size image

The ~ 20 kb and ~ 3 kb inversions (Inv1 and Inv2) of most Asteraceae were detected in all Saussurea plastomes (Fig. 1). Inv1 was located between the trnG-UCC and trnS-GCU genes; Inv2 (located between the trnS-GCU and trnE-UUC genes) was nested within the large inversion and shared one end-point with Inv1 (Fig. 1). Sliding window analysis showed much higher proportions of variable sites in single-copy regions than in the IR regions. Seven relatively highly variable regions (rps16-trnQ, trnS-trnC-petN, psbE-petL, ndhF-rpl32, rpl32-trnL, rps15 and ycf1) were identified from the plastome sequences (Fig. 3).

Selection analyses

Most protein coding genes showed a low dN/dS ratio (ω; Additional file 2: Figure S1), indicating that they have been under purifying selection. Only three genes (psbL, psbZ and ycf2) had ω > 1, but the branch model results revealed no significant difference between foreground and background branches (Table 2).

Table 2 Branch model results of three genes with ω > 1. ω = dN/dS, the ratio of nonsynonymous/synonymous substitution rates. np: number of parameters. ln L: log likelihood values. LRT: likelihood ratio test
Full size table

Phylogenetic analyses

Our phylogenomic analyses substantially increased resolution and provided robust backbone relationships of Saussurea (Fig. 4, Additional file 3: Figure S2). Characteristics of the three concatenated data sets are presented in Table 3. Dataset-3 had the highest number of parsimony-informative (PI) characters, followed by dataset-1 and dataset-2. Centaureinae were resolved as sister to Saussurea in datasets-1 and -3 with strong support, but not in dataset-2. All three datasets also strongly supported the monophyly of Saussurea (BS = 100), while three (Eriocoryne, Amphilaena, and Saussurea) of the four traditional subgenera were resolved as paraphyletic. Three main clades of Saussurea were identified. Clade 1 included three species of subg. Amphilaena (S. publifolia, S. sp. nov., S. involucrata), one of subg. Eriocoryne (S. lhozhagensis) and five of subg. Saussurea (S. durgae, S.przewalskii, S. salwinensis, S. delavayi, S. kingii). Clade 2 included two species of subg. Amphilaena (S. hookeri, S. obvalata) and four of subg. Eriocoryne (S. gnaphalodes, S. gossypiphora, S. psedoleucoma, and S. tridactyla). Clade 3 included two species of subg. Theodorea (S. japonica and S. tsoongii) and two Korean species (S. chabyoungsanica, S. polylepis). Datasets-1 and -2 resolved subg. Theodorea as sister to remaining Saussurea, whereas dataset-3 resolved clade 2 as sister to remaining Saussurea, albeit with low support (Fig. 4, Additional file 3: Figure S2). The coalescent-based result yielded an almost identical topology with the concatenation-based phylogeny (dataset-1), except for the position of S. kingii, which was resolved as sister to clade 1 + clade 2 (Additional file 4: Figure S3).

Table 3 Characteristics of the three different data sets
Full size table

Discussion

Plastome evolution

The 20 Saussurea plastomes in our analyses indicated that plastome evolution has been conservative within this genus. All Saussurea plastomes possessed the typical plastome structure of most Asteraceae, including both LSC inversions that are present in nearly all Asteraceae, as for example in Lactuca [27], Artemisia [29], Lasthenia [28], Taraxacum [39] and Mikania [40]. The expansion and contraction of the IR region has been demonstrated to be a significant source of length variation in some plastomes, e.g. early-diverging eudicots [41, 42] and Apiales [43]. In the present study, however, no significant IR length variation was detected among Saussurea plastomes (Fig. 2).

Fig. 2
figure 2

Comparison of the LSC, IR and SSC borders among seven Cardueae genera, with Lactuca sativa as a reference

Full size image

In our molecular evolutionary analysis, most protein-coding genes were found to be under purifying selection (Additional file 2: Figure S1). This pattern has also been demonstrated in other Asteraceae plastomes, such as in Mikania cordata [40] and Helianthus [44], reflecting the typically conservative evolution of plastome genes in green plants. Indeed, the best evidence for relaxation of purifying selection is in plants that have lost photosynthesis, in which genes involved directly in photosynthesis evolve much faster due to loss of function, typically resulting in pseudogenization and eventual gene loss [32, 34, 35]. Nevertheless, complete genome- and transcriptome-based analyses are necessary to fully investigate the importance of selection at protein-coding loci in plastids, given that most plastid proteins are encoded in the nucleus.

Phylogenetically informative sites

To resolve relationships among closely related species, it is imperative to identify rapidly evolving loci. Previous phylogenetic studies of Saussurea mainly favored three plastid loci (trnL-F, psbA-trnH, and matK) but these have failed to resolve relationships across the genus (e.g. [5,6,7,8]). Our analyses revealed relatively low nucleotide diversity in these three regions (Fig. 3), explaining the low resolution in previous analyses and highlighting the importance of exploring more of the plastome to obtain additional informative sites and regions. We found seven relatively variable regions: rps16-trnQ, trnS-trnC-petN, psbE-petL, ndhF-rpl32, rpl32-trnL, rps15 and ycf1. Of these, rps16-trnQ, trnC-petN, psbE-petL, rpl32- trnL, rps15 and ycf1 have been previously reported as hotspots of divergence and have been broadly used for reconstructing phylogeny in plant taxa [40, 45,46,47,48,49,50]. The lineage-specific, rapidly evolving regions identified here will facilitate further phylogenetic resolution of the large and diverse Saussurea.

Fig. 3
figure 3

Sliding window analysis of nucleotide variability (Pi) across 31 complete plastome sequences of Cardueae, with one copy of the IR included

Full size image

Phylogenetic relationships within Saussurea

The backbone relationships of Saussurea have been poorly resolved in previous molecular phylogenetic studies (e.g., [5,6,7,8, 10,11,12]). Our analyses greatly increased resolution with generally robust support (Fig. 4, Additional file 3: Figure S2). With the exception of subg. Theodora (the only monophyletic subgenus), there is relatively little concordance between the relationships recovered here and morphological characters used to define sections and subgenera [3, 4, 13]. In fact, these morphological characters have been shown to have adaptive value, as for example the dense woolly trichomes and colorful bracts that are used to circumscribe subg. Eriocoryne and subg. Amphilaena respectively. These two kinds of character states are prevalent among alpine species, and have been thought to protect plants from cold and UV-B radiation at high elevations [5, 51,52,53]. Hence, the discordance between phylogeny and morphology may reflect potential convergent evolution in Saussurea. It is also important to note that our estimate of phylogeny is based only on the plastome in a rapidly radiating group. Given that incomplete lineage sorting (ILS) or hybridization are most likely to obscure the species phylogeny among close relatives, it is possible that the addition of nuclear phylogenomic data may result in a different estimate of relationships in Saussurea. Consequently, it is essential to expand taxon and locus sampling significantly within Saussurea to better understand patterns of character state evolution and biogeography.

Fig. 4
figure 4

Inferred molecular phylogeny from ML (maximum likelihood) and BI (Bayesian inference) analyses using different data sets. a Complete plastome sequences (dataset-3); b All 79 CDS (dataset-1). Maximum likelihood bootstrap values (BS) and posterior probabilities (PP) are shown at nodes. Branches with * have 100% bootstrap support

Full size image

The clades formed by subg. Theodorea and sect. Laguranthera (S. durgae) were resolved as early-diverging groups in phylogenetic studies of Saussurea based on ITS and trnL-trnF [7]. In our concatenated datasets-1 and -2 and coalescent-based approach, the early-diverging position of subg. Theodorea was also supported, despite it being relatively distant phylogenetically from sect. Laguranthera. Across all concatenated datasets, S. kingii had the longest branch by far (Fig. 4, Additional file 3: Figure S2), which was also detected in the phylogenetic study of Wang and Liu [12]. As suggested there, this likely results from its biennial habit, as substitution rates are known to be higher in species with shorter generation times [54]. In addition, the systematic position of S. kingii was unstable between concatenated- and coalescent-based approaches, suggesting a further investigation may be required.

Incongruence at deeper levels among the trees resulting from our three concatenation-based analyses is likely related to differences in the number of parsimony-informative (PI) characters among data sets, with the highest number of PI characters in dataset-3 (Table 3). These differences likely explain the better overall support for the backbone of Saussurea in the tree based on dataset-3 (Fig. 4a) compared to the other trees. Given the relatively low taxonomic level (within a genus) of our study, it makes sense that including nucleotide sequence, especially for noncoding regions, would maximize the power to resolve relationships. We therefore recommend complete plastome data sets in these situations. The incongruence at a few backbone nodes is not surprising given how short these branches are; it is likely that few PI characters ever existed at these branches, and hence such nodes are sensitive to the conditions of phylogenetic analysis [23, 55].

Conclusions

Our analyses reveal both the structural conservatism and power of the plastome for resolving relationships in congeneric taxa. By examining signals of selection at protein-coding loci, we are able to eliminate systematic error due to selective biases as a source of topological incongruence. Hence, it is very likely that differences in topology among data sets are due primarily to differences in numbers of parsimony-informative characters. Our study further demonstrates that currently accepted subgeneric groups in Saussurea are likely based at least partly on convergent character states, and are therefore in need of revision. Moreover, greater taxon sampling is necessary to disentangle the patterns of character evolution and biogeography that are only hinted at here. Our results here provide helpful insight into which loci will provide the most PI sites in Saussurea and Cardueae, but they also suggest that complete plastome sequencing will be a valuable technique for resolving the relationships in this difficult genus.

Methods

Taxon sampling, chloroplast DNA isolation, high-throughput sequencing

We sequenced 17 new plastomes representing 16 currently described and one undescribed species of Saussurea; collection and voucher information are provided in Additional file 1: Table S1. These were added to the three previously reported plastomes available in GenBank (Additional file 1: Table S1). The circumscription and infrageneric treatment of Saussurea followed Flora of China and Flora of Pan-Himalaya [3, 4]. For all species, total DNA was extracted from fresh or silica gel-dried leaves with a modified CTAB (Cetyl trimethylammonium bromide) method [56]. Sequencing libraries were constructed and quantified following the methods introduced by Sun et al. [41]. For all plastomes, a 500-bp DNA TruSeq Illumina (Illumina Inc., San Diego, CA, USA) sequencing library was constructed using 2.5–5.0 ng sonicated DNA as input. Libraries were quantified using an Agilent 2100 Bioanalyzer (Agilent Technologies, Santa Clara, CA, USA) and by real-time quantitative PCR. Libraries were then multiplexed and sequenced using a 2 × 125 bp run on an Illumina HiSeq 2000 platform at Novogene in Kunming, Yunnan, China.

Plastome assembly, annotation, and comparative analyses

Raw sequence reads were subsequently filtered using Trimmomatic v.0.36 [57] with the following parameters: SLIDING WINDOW = 4:20, MINLEN = 50, LEADING = 3, TRAILING = 3, HEAD-CROP = 12, and AVGQUAL = 20. Remaining high-quality reads were assembled de novo into contigs with a minimum length of 1000 bp using CLC Genomics Workbench 11.0 (https://www.qiagenbioinformatics.com/) with default parameters. The resulting de novo contigs were then reference-assembled against the plastome of S. chabyoungsanica. Finished plastomes were annotated using DOGMA [58] and GeSeq [59]. Manual adjustments of start/stop codons and intron/exon boundaries were conducted in Geneious version 9.0.5 [60], using published plastomes of Saussurea as references. The tRNA genes were identified with tRNAscan-SE [61]. Physical maps of the circular plastomes were visualized with OGDRAW [62].

We performed plastome comparisons between Saussurea polylepis and six other Cardueae genera (Cirsium arvense, Carthamus tinctorius, Cynara cornigera, Centaurea diffusa, Silybum marianum, Atractylodes chinensis). All seven complete plastomes were aligned with ProgressiveMAUVE [63], assuming collinear genomes for the full alignment. To assess sequence divergence and determine highly phylogenetically informative sites, nucleotide variability (π) was calculated by sliding window analysis conducted in DnaSP version 6.11.01 [64] with all aligned plastome sequences of Saussurea. For the purposes of alignment, the SSC region was inverted manually in Geneious as necessary. The step size was set to 200 bp, with a 600 bp window length.

Phylogenetic analyses

Thirty-one taxa (Additional file 1: Table S1) of Cardueae (20 Saussurea + 11 outgroup genera from Cardueae) and two outgroup taxa of Cichorieae (Lactuca sativa, Taraxacum officinale) were included in phylogenetic analyses. Both concatenated and coalescent-based analyses were conducted. For concatenation-based approach, three datasets were analyzed: dataset-1 included the nucleotide sequences of all 79 protein-coding sequences (CDS); dataset-2 included the amino acid sequences of these 79 CDS; and dataset-3 included the complete plastome nucleotide sequences, including only one copy of the IR regions. Dataset-1 and -2 were created by concatenating alignments using PhyloSuite version 1.1.15 [65]. Characteristics of all three data sets were calculated using MEGA X [66]. For all concatenated data sets, Modeltest version 3.7 [67] was used to estimate the optimal model under the Akaike Information Criterion (AIC). Maximum likelihood (ML) analyses were conducted using RAxML version 8.2.10 [68] under the general time reversible model of nucleotide substitution, with the gamma model of rate heterogeneity (GTRGAMMA for dataset-1 and daset-3; PROTGAMMAAUTO for dataset-2). Bootstrap (BS) support was estimated with 1000 bootstrap replicates using the “rapid bootstrap” algorithm of RAxML. Bayesian inference (BI) was performed using MrBayes version 3.2.3 [69]. Two runs were conducted in parallel with four Markov chains (one cold and three heated), with each running for 5000,000 generations from a random starting tree and sampled every 5000 generations. Convergence was assessed by examining the average standard deviation of split frequencies (ASDF). After ASDF reached < 0.01, the first 25% of the trees were discarded as burn-in, and the remaining trees were used to construct majority-rule consensus trees.

For the coalescent-based analysis, ML unrooted trees for 79 CDS alignments were estimated separately using RAxML under the GTRGAMMA model with 500 bootstrap replicates. ASTRAL III version 5.6.2 algorithm [70] was used to estimate the species tree from 79 gene trees with node supports calculated as local posterior probabilities.

Analyses of signatures of selection

To test for evidence of selection in plastid protein coding genes, we estimated the ratio of nonsynonymous (dN) to synonymous (dS) substitutions (ω) for all 79 protein coding genes using CodeML in PAML version 4.9 [71] with the following settings: model = 0, seqtype =1, NSsites = 0. Genes showing higher ω were identified with the branch model [72, 73] to determine lineage-specific selection in plastomes of Saussurea. Following the recommendations in CodeML, the best ML tree determined by RAxML with dataset-1 using concatenation-based approach was used as the input topology, and the clade formed by Saussurea was set as a foreground branch. The likelihood ratio and P value were used to test if a model (“model = 2”) of positive selection on the foreground branch was a significant improvement over a null model (“model = 0”) where no positive selection occurred on the foreground branch.