Sampling and amplifications

A total of 88 taxa were sampled in the analysis representing all species of the two genera except for Gueldenstaedtia monophylla and Tibetia forrestii according to the most recent classification system by Zhu [5,22]. Gueldenstaedtia monophylla was recorded from Altai mountain area of Russia, Mongolia and China, but we are failed to find it in the field. As for T. forrestii, it might be a variant of T. yunnanensis as previously mentioned [8]. Five species of Chesneya were used in this study to test its relationship to Tibetia and Gueldenstaedtia. Outgroup taxa include samples of Caragana rosea, C. sinica, C. microphylla, C. korshinskii, C. arborescens, and Halimodendron halodendron, which are from the Caraganeae tribe, and Hedysarum vicioides was selected according to the previous molecular phylogeny [21,23]. Collection information is listed in S1 Table. Voucher specimens were deposited in the Kunming Institute of Botany Herbarium (KUN).

Total genomic DNA was isolated from leaf material, using the plant total DNA extraction Kit (BioTeke, Beijing, China) following the manufacturer’s protocol. Five molecular markers (nuclear ITS and plastid DNA sequences from rbcL, matK, trnL-F, and psbA-trnH) were sequenced. PCRs were performed in 25-uL reactions using 2 × Power Taq PCR MasterMix (BioTeke, Beijing, China). Typical reaction conditions were as follows: 4 min at 94°C for denaturation, followed by 32 cycles of 40 s at 94°C, 40 s at 55°C for annealing, 1 min 30 s at 72°C for primer extension, then a final 10 min incubation at 72°C. The amplified products were then purified using polyethylene glycol (PEG) precipitation using standard protocols. Cycle sequencing reactions were conducted using BigDye 3.1 reagents on an ABI 3730 automated sequencer (Applied Biosystems, Foster City, California, USA). All sequences are deposited in GenBank (S1 Table).

The original sequences were assembled into contigs and edited using the program Sequencher 4.14. Sequence alignment was performed using ClustalW version 1.8 [24] followed by manual modification in the program BioEdit v7.0.4 [25]. Phylogenetic trees were reconstructed using maximum parsimony (MP), maximum likelihood (ML), and Bayesian inference (BI). Maximum parsimony searches were performed with tree bisection-reconnection branch swapping, with MulTrees on, and simple taxon addition in PAUP* 4.0b10 [26]. Parsimony bootstrap (BP) support [27] for the clades was estimated as described above from 1000 heuristic search replicates, with 100 random taxon addition replicates saving all optimal trees at each step. The ML trees were inferred with RAxML BlackBox, online software based on the so-called rapid bootstrap [28]. Maximum likelihood was implemented starting from random trees, using 1000 rapid bootstrap inferences with RAxML GAMMA model of rate heterogeneity, and ML estimate of alpha-parameter GAMMA Model parameters in effect. Finally draw bootstrap support values on the best-scoring ML tree.

The optimal DNA substitution model was estimated using Akaike information criterion (AIC) in jModeltest 0.1.1 [29,30]. Bayesian inferences were implemented in MrBayes 3.1.2 [31] with models estimated as above. A Bayesian analysis was performed on the condition that the Bayesian Markov Chain Monte Carlo (MCMC) calculation is set to 10,000,000 generations with 4 incrementally heated chains and sampling every 1000 generations. The Tibetia species began and convergence estimations were graphically assessed using AWTY [32]. The remaining 9000 trees were sampled from the posterior distribution to calculate the posterior probabilities (PP).

The data were analyzed separately for ITS and for the combined chloroplast dataset. Congruence among the different datasets was first tested using the incongruence length difference (ILD) [33,34] test in PAUP. Taxa with missing data were excluded in the ILD test.

Divergence time estimation

Due to difficulties in aligning ITS sequences across the diversity of Papilionoideae taxa included in this study and the lack of a sufficient number of other sequences including rbcL, psbA-trnH, trnL-F from taxa representing all groups included in the analyses, matK was selected to estimate the origin and diversification time. Previous studies on the legume family have successfully used matK sequences for molecular dating [35,36]. Sequences from GenBank representing the main clades in the Papilionoideae were applied in the dating analysis, and Caesalpinia andamanica and Haematoxylum brasiletto were selected as outgroup taxa. A likelihood ratio test [37] was conducted for clock-like behavior. The results suggested that rate constancy in this data set is not supported at P < 0.05.

BEAST v1.8.0 (http://beast.bio.ed.ac.uk) was used to estimate divergence times [38], which employs a MCMC to co-estimate topology, substitution rates and node ages. All of the analyses were performed based on a General Time Reversible (GTR) nucleotide-substitution model with a Gamma distribution and four rate categories. Divergence times and the corresponding credibility intervals were estimated using a lognormal relaxed molecular clock model and the Birth-Death prior set. Posterior distributions of parameters were approximated using two independent MCMC analyses of 20,000,000 generations with a 10% burn-in. Samples from the two chains that yielded similar results were combined, and convergence of the chains was checked using the program Tracer 1.5 [39]. Samples from posterior distributions were summarized on a maximum clade credibility (MCC) tree with the maximum sum of posterior probabilities on its internal nodes using TreeAnnotator v1.6.1 with the posterior probability limit set to 0.5 to summarize mean node heights. The MCC tree was visualized using FigTree v1.3.1 and the means and 95% higher posterior densities (HPD) were obtained.

The age constraints derived from the legume fossil records were incorporated into the BEAST analysis. The origin time of Papilionoideae was set to 58.6 million years ago (Ma) according to Lavin et al. [35]. There are numerous options from the large fossil record from Papilionoideae, which have been dated to the Late Paleocene and come from America, Europe, Africa, and Asia [40]. Three fossil calibrations were used: (1) Robinia stem clade was fixed to 34 (± 0.1) Ma based on a wood fossil possessing apomorphic traits unequivocally related to the Robinia stem clade [41–44]; (2) the Diplotropis stem clade was set to 56 (± 0.1) Ma, based on a fossil of the leaves and pods that is from the Late Paleocene as well as Middle Eocene, and are very similar to Bowdichia [45,46]; (3) the Styphnolobium stem clade was set to a constraint of 40.1 (± 0.4) Ma (Middle Eocene), given that Styphnolobium and subgenus Cladrastis have a fossil of a pod and leaflets from Middle Eocene from Tennessee, USA [40].

Phylogenetic status of Tibetia and Gueldenstaedtia

The monophyly of Tibetia is robustly supported by molecular phylogenetic analyses based on the combined nuclear ITS and cpDNA dataset with a comprehensive sampling (PP = 1.00, BP = 100, Fig 2), in agreement with previous morphological and taxonomical studies [3,22]. Tibetia are endemic to the Hengduan Mountain region (i.e. Sino-Himalayan subkingdom), and its prostrate habit and dense glandular hair might be an adaptation to the high-altitude plateau.

Gueldenstaedtia is distinguished from Tibetia and other relatives with a chromosomal number of 2n = 14, and is also confirmed to be monophyletic by the molecular evidence (PP = 1.00, BP = 100; Fig 2). The genus is mostly restricted in the Sino-Japanese subkingdom (Fig 1) and distributed widely in eurytopic habitats, and has morphological characters with continuous variations that have caused challenges in the identification of infrageneric relationships due to the instability in these traits. The difficulty of distinguishing species morphologically for the genus is probably reflected in the difficulty of resolving infrageneric relationships based on molecular phylogenetic analyses.

The sister relationship of these two genera is also suggested in this study with high support (PP = 1.00, BP = 100, Fig 2). Morphologically, they share many synapomorphic features, such as small herbaceous plants, stems that are almost absent, or with very short or many branches, pinnate leaves, and racemes composed of several yellow or purple papilionaceous flowers, calyx base symmetric or suboblique, wings usually palmately nerved, legume valves not twisting, keel petals half as long as wings and style shorter than or as long as ovary [6,22,49]. The two genera are distributed throughout the Sino-Himalayan and Sino-Japanese subkingdoms in eastern Asia and flower from May to June. Based on this information, some scholars classified Tibetia and Gueldenstaedtia as one genus [50–53]. It is controversial to separate Tibetia from Gueldenstaedtia [4,50,52]; however, our phylogenetic results strongly support the monophyly of these genera. The two genera are also morphologically and cytologically distinct (Table 1). For example, Gueldenstaedtia always has lateral and separated stipules, non-spotted seeds, and chromosome base number of x = 7, while Tibetia has amplexicaul and united stiples, irregular spotted seeds, and chromosome base number of x = 8. Together with our phylogenetic results and the morphological and cytological characteristics, we support the separation of Tibetia as independent genus and as a sister group to Gueldenstaedtia.

Chesneya has been suggested to be one of the closest relatives of Tibetia and Gueldenstaedtia because of the cytological (chromosome number of Chesneya is x = 8) and molecular evidence [11,12,54]. Our data well recognize a clade including Gueldenstaedtia, Tibetia, and Chesneya species (PP = 1.00, BP = 99) and suggest Chesneya is sister to Tibetia and Gueldenstaedtia. This agreed with previous molecular analyses concerning these genera [10,12,55]. Furthermore, Chesneya species are perennial herbs with thick roots, lignification, very short stems, leaves imparipinnate, and similar pods to Gueldenstaedtia and Tibetia species, supporting their close relationship. Ranjbar and Karamian (2003) established the new tribe, Caraganeae, which joins Caragana, Halimodendron, Calophaca, Chesneya, and Gueldenstaedtia. This new tribe is morphologically characterized by the pedicel asymmetrically attached to a slightly gibbous ± tubular calyx (except Halimodendron), the valves of the pod generally twist upon dehiscence (except Halimodendron), and opening calyx during fruiting [13]. However, Duan et al. [10] did not support the monophyletic status of the tribe based on ITS, matK, trnL-F, and psbA-trnH sequences. This tribe is separated into two monophyletic subtribes, the Caraganean clade (including Caragana, Halimodendron, and Calophaca) and the Chesneyean clade (consisting of Chesneya, Gueldenstaedtia, and Tibetia).

Phylogenetic relationships within each genus

Two major lineages corresponding to flower color were well recognized in Tibetia (Fig 2). The first includes only T. tongolensis with yellow flowers and a glabrous ovary (clade A), which is distinct from all others in the genus, which have dark purple, violet, or blue flowers and a pubescent ovary (clade B). T. tongolensis is very unique with its yellow flowers in Tibetia. Additionally, the other species in Tibetia and their sister genus, Gueldenstaedtia, are all purple-flowered; therefore, the yellow flower color is apparently a derived character state in this group.

Three lineages were well distinguished in the clade B, and relationships among the three lineages are unclear (Fig 2). The first includes T. yunnanensis and T. himalaica from sect. Tibetia which is morphologically characterized by 4-colporate pollen grains and stipules acute at the apex. However, some individual samples from T. yunnanensis and T. himalaica appear within different clades on the plastid or nuclear tree (S1 and S2 Figs), which may be caused by DNA mutation or random introgression events. Another clade includes only T. yadongensis with 3-cloporate pollen grains, belonging to sect. Glabrae [22,49]. The third lineage includes only T. liangshanensis, which was originally treated as T. forrestii (Ali) P. C. Li in Zhu [22] However, phylogenetic and morphological evidences supported the recognition of T. liangshanensis as a different species from T. yunnanensis [8], which has significantly different morphological characters, including more leaflets per pinnate leaf, a smooth abaxial epidermis, and free stipules, vs. one to four leaflets per pinnate leaf, a squamose abaxial epidermis, and stipules connecting at the base in T. yunnanensis [8].

Although the monophyly of Gueldenstaedtia is robustly supported, the infrageneric relationships are uncertain as shown with extremely short branch lengths and low resolutions (Fig 2). This might be a result of the recent origin of Gueldenstaedtia, without enough time for sequence variation to accumulate. The circumscription of species within Gueldenstaedtia is usually based on whether the stem is lignified or not, the leaflet is broad or narrow, the flowers are many or few, and the plants pubescent or glabrous [4]. However, these characteristics are not dependable to discriminate different species that may have continuous morphological variation. It is possible to consider Gueldenstaedtia as a species complex with abundant continuous characters.

Biogeographic differentiation pattern within eastern Asia

The divergence of Tibetia from Gueldenstaedtia was estimated at about 17.23 Ma (95% HPD: 9.61–25.52 Ma) in the early Miocene. Many other plant groups from the QTP and adjacent regions have also been estimated to have originated or diverged from their relatives around this time. For example, the diversification of the cushion-like “Androsace group” in the QTP from other sister genera was speculated at 18.02 Ma [56]. The explosive radiation of the Ligularia–Cremanthodium—Parasenecio complex also occurred mostly within the last 20 Ma likewise, perhaps effected by the major uplifts of the QTP in the early Miocene [57]. Most species of the temperate Asian genus Caragana are distributed on the QTP and in northwestern China. Caragana species diversified at about 16–14 Ma, and were also presumably triggered by the QTP uplift at 21–17 Ma [58]. The Sino-Himalayan endemic genus Cyananthus, with an estimated origin 23–12 Ma, may be another case of diversification in response to the Himalayan uplifts [59]. The QTP went through roughly four intense uplifts in total, i.e., 25–17 Ma, 15–13 Ma, 8–7 Ma, and 3.5–1.6 Ma [60–63]. The Himalayan uplift around 25–17 Ma was one of the most significant events [64], which is considered to have triggered the diversification of many plant and animal taxa in this area [57,58,65,66]. Similar to the taxa discussed above that originated and diversified during the QTP uplift, the split of Tibetia and Gueldenstaedtia that occurred about 17 Ma might be also have been driven by this extensive architectonical movement during the early Miocene.

The flora of eastern Asia is subdivided into the Sino-Himalayan and Sino-Japanese subkingdoms [16,17]. The biogeographic pattern of Tibetia and Gueldenstaedtia species mostly corresponds to these two subkingdoms. The Sino-Himalayan Tibetia shows a much higher level of infrageneric divergence than the Sino-Japanese Gueldenstaedtia (Table 2). Tibetia diversified in the Sino-Himalayan region at ca. 5.25 Ma in late Miocene, which has subsequently undergone a significant change in topography and climate corresponding to the phased uplifts of the Himalayas [67]. This region is characterized by rich biodiversity, with many evolutionary radiations, and diversification of many plant groups associated with the varied isolated habitats and mountains [68–70]. In contrast, Gueldenstaedtia exhibits very low morphological variation and genetic differentiation. The genus is distributed in northern and central China all the way to Siberia, a region dominated by vast plains, which has rarely been affected by geological events. The physical environment has remained relatively stable and homogenous with fewer geographical barriers and weaker divergent selection pressures, which could explain the low molecular differentiation for Gueldenstaedtia.