Our sampling was designed largely following the generic demarcations in Flora Reipublicae Popularis Sinicae (Liou 1993; Li 1993; Cui 1998). We included 101 accessions, covering 97 species, containing 39 species of Caraganinae (represented by Calophaca, Halimodendron and all 5 sections of Caragana according to Zhang 1997) and 40 accessions (36 species) of Chesneyinae (including Chesneya, Chesniella, Gueldenstaedtia and Tibetia, tentatively treating Spongiocarpella in Chesneya, which were more comprehensively sampled than previous studies [Ranjbar et al. 2014; Duan et al. 2015; Zhang et al. 2015b]). 82 new sequences were generated in this work.

To better resolve the relationships of subtribes Caraganinae and Chesneyinae, 11 Galegeae species (8 genera) and 5 Hedysareae species (4 genera) were also sampled. Cicer microphyllum Royle ex Bentham, Dalbergia hupeana Hance, Lathyrus latifolius L., Robinia pseudoacacia L., Trifolium repens L. and Wisteria sinensis (Sims) Sweet were selected as outgroups based on previous studies (Wojciechowski et al. 2000, 2004; Wojciechowski 2003). Sequences of 40 accessions (representing 40 species) were downloaded from GenBank (see Suppl. material 1 for details). Most accessions we sampled were collected from the field or herbarium specimens. Onobrychis arenaria DC. was obtained from seedlings germinated from seeds provided by the Royal Botanic Gardens, Kew.

Total genomic DNAs were extracted from silica-gel dried leaves or herbarium material using the Plant DNA Extraction Kit - AGP965/960 (AutoGen, Holliston, MA, USA) or the DNeasy Plant Mini Kit (Qiagen, Valencia, USA). Polymerase chain reactions (PCR) were prepared in 25µL containing 1.5 mM MgCl₂, 0.2 mM of each dNTP, 0.4 mM of each primer, 1 U of Taq polymerase (Bioline, Aberdeen, Scotland, UK), and using 10–50 ng (2.5 µL) template DNAs, following Wen et al. (2007). The PCRs for ITS (primer pair: ITS4 and ITS5a) and psbA-trnH (primer pair: psbA and trnH) were performed according to Stanford et al. (2000) and Hamilton (1999), respectively. The PCR primer pair for trnL-F was “c” and “f” as in Zhu et al. (2013) and Taberlet et al. (1991), and the thermal cycling program followed Soejima and Wen (2006). The barcoding region of the matK marker was amplified and sequenced with the primer pair Kim-3F/Kim-1R (CBOL Plant Working Group 2009; China Plant BOL Group 2011), and the amplification conditions were: 95°C (5min) for DNA pre-denaturation; 94°C (40s), 48°C (40s) and 72°C (100s) for 35 cycles; 72°C (10min) for final extension. PCR products were cleaned using ExoSAP-IT (cat. # 78201, USB Corporation, Cleveland, OH, USA) following the manufacturer’s instruction. Purified products were sequenced from both directions with BigDye 3.1 reagents on an ABI 3730 automated sequencer (Applied Biosystems, Foster City, CA, USA).

Sequences were assembled with Geneious 7.1 (http://www.geneious.com/), and aligned using MUSCLE 3.8.31 (Edgar 2004), followed by manual adjustments in Geneious 7.1. Because the chloroplast markers putatively evolve as a single molecule, sequences of the three plastid markers (matK, trnL-F and psbA-trnH) were directly concatenated. Topological discordance was investigated by comparing the ITS and the concatenated plastid trees (as in García et al. 2014). To further determine the compatibility between these two datasets, an incongruence length difference (ILD) test and an approximately unbiased (AU) test were conducted with PAUP* (Swofford 2003) and CONSEL (Shimodaira and Hasegawa 2001; using site-wise likelihood values estimated by RA×ML; Stamatakis et al. 2008) programs, respectively. The tests retrieved the p values of 0.01 and 0.0001, respectively, suggesting that the incongruence between these two datasets was significant. The ITS and the concatenated plastid sequences were thus analyzed separately.

Phylogenetic analyses were carried out using Bayesian inference (BI; Rannala and Yang 1996; Mau et al. 1999) with MrBayes 3.2.5 (Ronquist and Huelsenbeck 2003; Ronquist et al. 2012). Nucleotide substitution model parameters were determined prior to BI using the corrected Akaike information criterion (AIC) in jModeltest 2.1.7. (Posada 2008; Darriba et al. 2012). For the ITS dataset, boundaries of the 5.8S region to the ITS1 and the ITS2 regions were determined by comparison with the published 5.8S sequence of Vicia faba L. (Nazar and Wildeman 1981; Yokota et al. 1989), and the sequence substitution models for the ITS1, 5.8S and ITS2 regions were determined separately. Similarly, the models for each of the three plastid markers were estimated for the best-fit models, which were used in the BI analysis for concatenated plastid sequences in a partitioned scheme.

In the BI, the Markov chain Monte Carlo (MCMC) search was run by two replicates for 10,000,000 generations, sampling one tree every 1,000 generations. After the first 2,500,000 generations (2,500 trees) were discarded as burn-in, a 50% majority-rule consensus tree and posterior probabilities were obtained among the remaining trees. Results were checked using the program Tracer 1.5 (Rambaut and Drummond 2007) to ensure that plots of the two runs were converging and the value of the effective sample size for each replicate was above 200. Maximum likelihood (ML) analyses were conducted using RAxML-MPI v8.2 (Stamatakis 2014) with dataset partition scheme the same as in the BI and the following settings: rapid bootstrap analysis with 1,000 replicates and search for best-scoring ML tree in one program run, starting with a random seed, selecting the GTR model. Bootstrap values (LBS), as well as posterior probabilities (PP) were labeled on the corresponding branches of the Bayesian trees.

In the ITS tree (Fig. 1), the Astragalean clade (PP = 1, LBS = 100%; including Astragalus L., Colutea L., Eremosparton Fisch. & C.A.Mey., Lessertia R.Br. ex W.T.Aiton, Oxytropis DC., and Swainsona Salisb.), the Vicioid clade (PP = 1, LBS = 100%; represented by Trifolium, Lathyrus, Cicer and Galega L.), tribe Hedysareae (PP = 1, LBS = 98%), subtribes Caraganinae (PP = 1, LBS = 98%) and Chesneyinae (PP = 1, LBS = 100%) were each strongly supported.

Subtribe Caraganinae contained three genera, within which Calophaca was monophyletic (PP = 1, LBS = 96%), but Calophaca and Halimodendron were embedded within the paraphyletic Caragana. Within subtribe Chesneyinae, Gueldenstaedtia (PP = 1, LBS = 100%) and Tibetia (PP = 1, LBS = 100%) were each monophyletic and together they formed a clade (the GUT clade, shown in blue; PP = 1, LBS = 100%). Two accessions of former Chesneya macrantha Cheng f. ex H.C.Fu constituted a robustly supported branch nested in a monophyletic Chesniella (displayed in green; PP = 0.98, LBS = 89%), while other accessions of Chesneya formed another clade (Chesneyas.s.; shown in red; PP = 1, LBS = 100%; Fig. 1), which contained three well-supported sections (details see Discussion; PP = 1 & LBS = 100%, PP = 0.98 & LBS = 96% and PP = 1 & LBS = 100%, respectively).

Similar to the ITS results, the plastid tree (Fig. 2) also showed the monophyly of both subtribes Caraganinae (PP = 1, LBS = 100%) and Chesneyinae (PP = 1, LBS = 100%). Calophaca and Halimodendron were nested in Caragana in different places from the ITS tree, but such placement was weakly supported. Caragana also showed its paraphyly, with C. sect. Bracteolatae (Kom.) M.L.Zhang (PP = 1, LBS = 100%), C. sect. Caragana Kom. (PP = 1, LBS = 98%), C. sect. Frutescentes (Kom.) Sanchir (PP = 1, LBS = 98%) and C. sect. Spinosae (Kom.) Y.Z.Zhao (PP = 1, LBS = 100%) each strongly supported. Unlike in the ITS tree, Chesneyas.s. and Chesniella were sisters in the plastid tree (PP = 1, LBS = 92%; Fig. 2). As in the ITS tree, the GUT clade (PP = 1, LBS = 100%) contained Gueldenstaedtia (PP = 1, LBS = 100%) and Tibetia (PP = 1, LBS = 100%), with each genus being monophyletic.

Within Caraganinae, Halimodendron contains only H. halodendron (Pall.) Druce with its distribution roughly overlapping with that of Calophaca (Lock 2005). This species is morphologically unique in Caraganinae with its inflated pods (Gorshkova 1945; Liu et al. 2010b). Consistent with previous studies (Zhang et al. 2009; Zhang and Fritsch 2010), our results also showed that Halimodendron is nested within Caragana. The phylogenetic evidence hence supports treating Halimodendron as a section within Caragana, i.e., Caragana sect. Halimodenron (Fisch. ex DC.) L.Duan, J.Wen & Zhao Y.Chang. We also resurrect the name Caragana halodendron (Pallas) Dumont de Courset based on Halimodendron halodendron (Figs 1, 2; see Taxonomic Treatment).

Calophaca morphologically resembles Caragana, and it is only distinguished from the latter by its imparipinnate leaves, rachises without thorns, and relatively denser racemes (Borissova 1945; Liu et al. 2010b). Calophaca contains 5–8 species mainly distributed in mountainous areas of central Asia, with one species extending to eastern Europe, and one endemic to northern China (Borissova 1945; Tutin et al. 1968; Yakovlev et al. 1996; Lock 2005; Liu et al. 2010b; Zhang et al. 2015a). The embedded position of Calophaca within Caragana argues that its classification needs to be placed in the broader phylogenetic framework of Caragana, which is supported by our results (Figs 1, 2) and several previous studies (e.g., Zhang et al. 2009, 2010, 2015a, b; Duan et al. 2015). We thus merge Calophaca into Caragana and recognize it at the sectional level as Caragana sect. Calophaca (Fisch. ex DC.) L.Duan, J.Wen & Zhao Y.Chang (see Taxonomic Treatment). The species-level nomenclatural changes will be made in a follow-up paper.

The taxonomy of Caragana has been investigated by various authors (Komarov 1908; Poyarkova 1945; Moore 1968; Sanczir 1979, 2000; Gorbunova 1984; Zhao 1993; Zhou 1996; Zhang 1997; Sanchir 1999; Chang 2008). However, Caraganas.s. as previously circumscribed is clearly paraphylytic (Zhang et al. 2009; Duan et al. 2015). We herein propose the delimitation of Caraganas.l. to ensure the generic monophyly (see Taxonomic Treatment). Caragana as defined now contains taxa of Calophaca, former Caraganas.s. and Halimodendron (Figs 1, 2), which is classified into seven sections: Car. sect. Bracteolatae M.L.Zhang, Car. sect. Calophaca, Car. sect. Caragana, Car. sect. Frutescentes (Kom.) Sancz., Car. sect. Halimodenron, Car. sect. Jabatae (Kom.) Y.Z.Zhao and Car. sect. Spinosae (Kom.) Y.Z.Zhao. Although Caraganas.l. is morphologically diverse, this genus can be diagnosed by its shrubby habit, saccate, oblique calyx bases, pinnate nerves on the wing petals and twisted, dehiscent pods (except for Car. holodendron). The expanded concept of Caragana is also supported by cytological evidence (Moore 1968; Chang 1993; Li 1993; Zhou et al. 2002; Chang 2008): most xeric and psychric taxa of Caraganas.l. have the same basic chromosome number (x = 8).

At the sectional level, our ITS tree (Fig. 1) indicated a strongly supported Car. sect. Calophaca. On the other hand, former Caraganas.s. was divided into five sections mainly based on the combinations of leaf (pinnate or digitate) and petiole/rachis (persistent or caducous) characters (Zhang 1997). Three main sections, Car. sect. Bracteolatae, Car. sect. Caragana and Car. sect. Frutescentes, evolved likely accompanying the rapid uplifts of the Qinghai-Tibet Plateau (QTP) at around 8 Ma (Zhang et al. 2009). These three sections also largely correspond to psychrophytic, mesophytic and xerophytic habitats, respectively (Zhang and Fritsch 2010). Our analyses supported the monophyly of the three sections, with Car. sect. Frutescentes only being monophyletic in the plastid tree (also see Zhang et al. 2009; Duan et al. 2015; and see below for an exceptional case in Car. sect. Frutescentes). Our ITS results failed to resolve a monophyletic Car. sect. Frutescentes (Fig. 1), but this may be due to insufficient informative sites in the ITS data. Furthermore, we only sampled one series for Car. sect. Spinosae (Car. ser. Spinosae Kom.), thus cannot assess its monophyly (Figs 1, 2). Caragana sect. Jabatae was suggested to have experienced a rapid radiation at 3.4–1.8 Ma (Zhang and Fritsch 2010), which may partly explain its poorly resolved relationships in our trees (Figs 1, 2; also see Zhang et al. 2009; Duan et al. 2015).

At the infra-sectional level, Car. ser. Bracteolatae Kom. and Car. ser. Spinosae are well-supported by our results (not labeled in the trees). Our results are therefore not completely congruent with Zhang et al. (2009), possibly due to differences in taxon sampling. Interestingly, a strongly supported psychric group is found within the mainly xeric section Car. sect. Frutescentes (Zhao 2009). This group is represented by Car. brevifolia Kom., Car. chinghaiensis Y.X.Liou, Car. densa Kom. and Car. versicolor Benth. (in Fig. 1; but weakly supported in the plastid tree). Most species of Car. sect. Frutescentes range from eastern Europe to northern China, Mongolia and Siberia, however, this above-mentioned psychric group is distributed in the southern edge of northern China, extending to Tibet and its neighboring regions. It may represent a vicariant transitional group of Car. sect. Bracteolatae, Car. sect. Jubatae pro parte, Car. sect. Spinosae pro parte (psychrophytic habitat) and Car. sect. Frutescentes. Other cases of vicariant distributions have been noted in Caragana, and vicariance was considered as an important biogeographic pattern for this genus. For example, three closely related species in Car. sect. Caragana, Car. microphylla Lam., Car. intermedia Kuang & H.C.Fu and Car. korshinskii Kom., show non-overlapping to only slightly overlapping distributions in northeast to northwest China (Shue and Hao 1989; Zhang and Wang 1993; Zhang 1998; Chang 2008).

The subtribe Chesneyinae, as established by Ranjbar et al. (2014), was supported to be monophyletic in our trees (Figs 1, 2). Three main clades can be recognized within this subtribe: the GUT clade, Chesneyas.s. and Chesniella (Figs 1, 2).

This subtribe contains ca. 50 species and differs from the Astragalean clade by twisted valves (e.g., in Chesneya), but a few species of Astragalus also have twisted legumes. Taxa of Chesneyinae are distinguished from Hedysareae by their dehiscent pods (Borissova 1945; Yakovlev et al. 1996; Liu et al. 2010a). The genera of Chesneyinae are distributed in central and eastern Asia, Tibet, Mongolia and Siberia, extending to eastern Turkey and Armenia (Fig. 3A; Borissova 1945; Davis 1970; Rechinger 1984; Lock and Schrire 2005; Liu et al. 2010a), which are largely adapted to xerophytic (Chesneya and Chesniella), mesophytic (Gueldenstaedtia) and psychrophytic (Tibetia) habitats, respectively, although some species of Chesneya (see discussion below) and a few of Gueldenstaedtia are psychric taxa. The uplift of the QTP and aridification of the former Tethys region might have driven the origination and divergence of genera in the subtribe Chesneyinae (Wen et al. 2014; Meng et al. 2015; Zhang et al. 2015b).

Figure 3. 

Distribution (A) and representative plants (B–H) of genera in Chesneyinae. A red – Chesneya, green – Chesniella, blue – Gueldenstaedtia and yellow – Tibetia B Chesneya acaulis C Chesneya spinosa D Chesneya nubigena E Chesniella macrantha F Chesniella ferganensis G Gueldenstaedtia verna H Tibetia yadongensis.

The ITS and plastid topologies are incongruent within Chesneyinae. Chesneyas.s. formed a clade with the GUT clade in the ITS tree (Fig. 1), whereas it was sister to Chesniella in the plastid tree (Fig. 2). Both relationships were well-supported. Various mechanisms have been proposed to explain discordant topologies between gene trees, such as allopolyploidy, hybridization, horizontal gene transfer, incomplete lineage sorting (ILS), different rate of molecular evolution, and chloroplast capture (Degtjareva et al. 2012; García et al. 2014; Yi et al. 2015).

Allopolyploidy can be ruled out for two reasons. First, taxa within Chesneyinae are diploid (Nie et al. 2002; Yang 2002; Sepet et al. 2014), with no evidence of polyploidy in this subtribe and its allied tribes. Second, deep lineages of Chesneyinae basically display a consistent chromosome number (x = 8; Nie et al. 2002; Sepet et al. 2014), with the only exception of Gueldenstaedtia (x = 7; Yang 2002), which has relatively recently diverged (ca. 15.23 Ma; Zhang et al. 2015b).

ILS and chloroplast capture seem more likely mechanisms for the present case (Tsitrone et al. 2003; Deng et al. 2015; Sun et al. 2015). A time-calibrated phylogeny may facilitate the exploration of the likely mechanism. Incomplete lineage sorting, which rarely occurs in deep lineage (Sun et al. 2015), prevails with bifurcation patterns of the shallow lineages of gene trees (especially at the specific level; Xu et al. 2012), and usually takes place in groups with relatively recent diversification times (García et al. 2014). Zhang et al. (2015b) estimated that the main clades of subtribe Chesneyinae split at ca. 28 Ma, which is beyond the time frame supporting ILS of ancestral polymorphisms (as suggested by Xu et al. 2012). On the other hand, biogeographic patterns can also be taken into consideration (Goodman et al. 1999). Given peripatry and parapatry may have been involved in the evolution of Chesneyinae, if ILS occurred, the main clades would hardly be resolved with well-supported dichotomy as presented herein. Hence, although ILS could not be completely excluded in this case, we regarded chloroplast capture as the most likely cause for the discordant position of Chesneyas.s.

Compared to the biparental inheritance of the nuclear genome, plastid DNA of angiosperms is usually uniparentally transmitted, especially maternally (Corriveau and Coleman 1988; McCauley et al. 2007; Wicke et al. 2011). Nevertheless, the plastid DNA of the inverted repeat lacking clade (IRLC; see Figs 1, 2; also as in Lavin et al. 1990; Wojciechowski et al. 2000) in Leguminosae was reported to be inherited paternally or biparentally (Zhang et al. 2003), confirmed by cytoplasmic and phylogenetic studies focusing on Medicago L. (paternal transmission; Schumann and Hancock 1989; Masoud et al. 1990; Havananda et al. 2010) and Wisteria Nutt. (Hu et al. 2005; Trusty et al. 2007). As Chesneyas.s. belongs to IRLC, a paternal inheritance scenario might be the case for the plastid DNA of Chesneyas.s.

We herein hypothesize a chloroplast capture event in the origin of Chesneyas.s. as follows. The common ancestor of Chesniella served as the putative paternal parent of Chesneyas.s. (sister to Chesneyas.s. in the plastid tree; Fig. 2). The maternal parent most likely was the common ancestor of the GUT clade. Their hybrids, with plastid from the paternal parent, may have continuously backcrossed with the maternal parent, and led to Chesneyas.s. inheriting most of the nuclear genome maternally (Fig. 1). Such a chloroplast capture event via introgression likely took place in the Miocene, because the divergence of Chesneyas.s. was dated to be 16.56 Ma and that of Chesniella was estimated as 19.81 Ma (Zhang et al. 2015b).

Analyses of Zhang et al. (2015b) revealed that the divergence of Chesneya and Chesniella most likely occurred around the QTP. Our analysis further indicated the psychric group of Chesneya diverged first in this genus (C. sect. Pulvinatae, see Discussion below). It is probable that the common ancestor of Chesniella adapted to psychrophytic habitats. However, the extant Chesniella is rarely distributed on the QTP. As for the GUT clade, Gueldenstaedtia possesses a unique chromosome number (x = 7; Yang 2002) within the subtribe. Most species of Gueldenstaedtia are adapted to mesophytic habitats of temperate northern and eastern Asia (Fig. 3A), in contrast to the rest of Chesneyinae, which are psychric or xeric taxa. Such a correlation among the variation of chromosome numbers and adaptation to different habitats has also been recorded in other taxa, such as Hedysarum (Tang 2005; Duan et al. 2015), Passiflora (Hansen et al. 2006) and Amaryllidaceae (García et al. 2014). But the mechanisms of these types of adaptation need to be further explored with robust phylogenetic, ecological and biogeographic analyses in our future efforts.

Chesneya is the type genus of Chesneyinae, with ca. 35 species (see Fig. 3B–D). This genus has its distribution from the Himalayan region to northwestern China and Mongolia, through central and western Asia, westward to Turkey and Armenia (Fig. 3A; Borissova 1945; Davis 1970; Yakovlev et al. 1996; Lock and Schrire 2005; Fig. 3A). Our results suggest that the formerly circumscribed Chesneya, which contains two well-supported but separated parts: the core Chesneyas.s. and the outlier C. macrantha (Fig. 3E) (as in Li 1993 & Zhu and Larsen 2010), is not monophyletic (Figs 1, 2). Chesneya spinosa P.C.Li (Fig. 3C) of Chesneyas.s. is morphologically similar to C. macrantha (Li 1981). However, C. spinosa is distributed in southern Tibet, while C. macrantha is restricted to the dry lands of Mongolia and northwestern China (Li and Ni 1985; Fu 1989). They occupy psychrophytic and xerophytic habitats, respectively, and are clearly not sister to each other (Figs 1, 2).

Chesneya macrantha is nested within a monophyletic Chesniella according to our ITS tree (Fig. 1), and in the plastid tree, it is sister to the type of Chesniella: Ch. ferganensis (Korsh.) Boriss. (Borissova 1964; see Fig. 2, 3F). Chesneya macrantha shows some distinct morphologies from the other species in Chesniella, including its pulvinate habit and persistent leaf rachis (Li 1993), but this species generally share distribution areas, xerophytic habitats, and some synapomorphies, such as membranous stipules, hairy standard and ovate leaflets with cuneate apices, with Chesniella (Li and Ni 1985; Fu 1989; Zhu and Larsen 2010). Therefore, the transfer of Chesneya macrantha to Chesniella is supported by morphological, geographic and phylogenetic evidence (see Taxonomic Treatment). On the other hand, Chesneya was thus re-delimited based on the monophyletic Chesneyas.s.

After its establishment by Lindley (1839), Chesneya was divided into C. sect. Macrocarpon Boriss. and C. sect. Microcarpon Boriss. mainly based on pod morphology (Borissova 1945). The latter was segregated as the genus Chesniella by Borissova (1964), and this treatment was followed by Li (1993) and Zhu and Larsen (2010). Zhang et al. (2015b) informally classified Chesneya into five sections without detailed taxonomic treatment. Not all their sections were monophylytic, and the diagnostic characters and distributions of several sections were overlapping to some extent.

The presently demarcated Chesneya was assigned into three strongly supported sections herein (as in the key of Chesneya proposed by Li 1993; details see Figs 1, 2 and Taxonomic Treatment). Chesneya sect. Macrocarpon possesses non-pulvinate habit, reduced stems, truncate or emarginate leaflet apices and caducous petiole and rachis (Borissova 1945). This section is composed of most species of Chesneya, including the type species: C. rytidosperma Jaub. et Spach (see Fig. 2; Borissova 1945; Davis 1970; Rechinger 1984). Chesneya sect. Macrocarpon was thus treated as C. sect. Chesneya (Fig. 3B). Unlike this section, petioles and rachises of C. sect. Pulvinatae M.L.Zhang (Zhang et al., 2015b; see Fig. 3D) are persistent and pubescent. However, most species in C. sect. Pulvinatae have blackened and curved petioles and rachises, while those of one of its species, C. spinosa, are hardened and spiny. Besides, C. spinosa formed a clade separated from C. sect. Pulvinatae. Hence, it is appropriate to segregate this species to form a new monotypic section: C. sect. Spinosae L.Duan, J.Wen & Zhao Y.Chang (see see Fig. 3C and Taxonomic treatment).

The infra-sectional relationships within C. sect. Chesneya are basically unresolved in our ITS trees (Fig. 1), and this section is undersampled in the plastid trees (Fig. 2). As for C. sect. Pulvinatae, two accessions of C. nubigena (D.Don) Ali formed a clade, being sister to C. purpurea P.C.Li (Figs 1, 2). Based on such well-supported tree topologies and several morphological differences, such as smaller leaflets and purple corollae, the specific status of C. purpurea was retained herein (as in Li 1981, 1993).

The xeric C. sect. Chesneya grows on dry slopes or desert margins of northwestern China, Mongolia and central Asia (see Fig. 3B; Borissova 1945; Rechinger 1984; Lock and Simpson 1991; Yakovlev et al. 1996; Zhu and Larsen 2010). This section is morphologically similar to Chesniella (Fig. 3F) and their distributions are more or less overlapping (Borissova 1945; Li, 1993), whereas they are not phylogenetically close to each other (Figs 1, 2). Such a phenomenon may be due to convergent evolution (Degtjareva et al. 2012). Chesneya sect. Spinosae (Fig. 3C) and C. sect. Pulvinatae (Fig. 3D) are restricted to Tibet and adjacent regions, adapting to high-altitude psychrophytic habitats (Ali 1977; Zhu and Larsen 2010). The evolutionary history of Chesneya appears complex, whereas the elevation of the QTP and the subsequent aridifications may have played an important role (Meng et al. 2015; Zhang et al. 2015b), as in former Calophaca (Zhang et al. 2015a), Caragana (Zhang and Fritsch 2010) and Hedysarum (Shue 1985; Duan et al. 2015).

Most previous workers did not accept the generic status of Chesniella, treating it within Chesneya (Borissova 1945; Li 1981; Rechinger 1984; Zhu and Cao 1986; Fu 1987, 1989; Yakovlev 1988; Yakovlev et al. 1991). Nevertheless, Li (1993) and Zhu and Larsen (2010) stated that the former is distinguishable from the latter by non-reduced stems, membranous stipules, obviously smaller calyxes, flowers and pods. With the inclusion of Ch. macrantha (Fig. 3E), our results justified the monophyly of Chesniella (Figs 1, 2), consistent with Zhang et al. (2015b). Within Chesniella, two well-supported groups were resolved in our ITS tree (Fig. 1). Chesniella macrantha and Ch. mongolica (Maxim.) Boriss. constituted group A, the group B included Ch. ferganensis, Ch. gracilis Boriss. and Ch. tribuloides (Nevski.) Boriss. The former confined in Mongolia and Inner Mongolia of China, to the contrast, the latter ranged from northwestern China to central Asia, which implied vicariance caused by Altai Mountain may drive the divergence of these two groups. However, due to undersampling and distinct morphology of Ch. macrantha in Chesniella, the evolution history and infrageneric taxonomy of this genus needs to be further explored.

Yakovlev and Sviazeva (1987) erected Spongiocarpella as a segregate genus from Chesneya in the light of the former’s spongiose legumes. Such treatment was followed by Yakovlev (1988), Fu (1989) and Yakovlev et al. (1996), but was rejected by Li (1993), Zhu (1996), Qian (1998) and Zhu and Larsen (2010). Based on field and herbarium studies, we concur with Zhu (1996) that the sponge-like pericarp is an unstable character. Additionally, several species formerly assigned to Spongiocarpella were represented in our study, including Chesneya nubigena (D.Don) Ali, C. Spinosa and Chesniella macrantha. They did not form a monophyletic group (Figs 1, 2). Thus, our data do not support the generic status of Spongiocarpella (as in Zhu 1996; Zhu and Larsen 2010; Ranjbar et al. 2014; Zhang et al. 2015b).

Gueldenstaedtia is a small genus comprised of ca. 10 species and is distinguished from Chesneya by its palmately nerved wing petals (vs. pinnately in Chesneya) and non-twisted pod valves (vs. twisted) (see Fig. 3G; Liu et al. 2010a). This genus ranges from the Sino-Himalayan region to Mongolia and Siberia (Lock and Schrire 2005; see Fig. 3A). It was established by Fischer (1823) and revised by Fedtschenko (1927), Jacot (1927) and Kitagawa (1936). Ali (1962) divided it into G. subg. Gueldenstaedtia and G. subg. Tibetia Ali, but the latter was elevated to the generic rank by Tsui (1979) based on characters of stems, stipules, styles and seeds (see Fig. 3H). The genus Tibetia was generally accepted in subsequent revisions (Shue 1992; Yakovlev et al. 1996; Cui 1998; Wu 1999; Zhu 2004, 2005a; Bao and Brach 2010), and it is confined to Tibet and the adjacent regions including southern Gansu, southern Qinghai, western Sichuan and northwestern Yunnan of China, northern India, Nepal and Buhtan (Tsui 1979; Grierson and Long 1987; Lock and Schrire 2005; Zhu 2005a; Bao and Brach 2010).

Gueldenstaedtia and Tibetia were each supported to be monophyletic, and the two genera together form the GUT clade (Figs 1, 2). It seems valid to retain the generic status of each genus, which is also supported by karyological studies (Nie et al. 2002; Yang 2002; Zhu 2005b): Gueldenstaedtia (x = 7) vs. Tibetia (x = 8). Within Gueldenstaedtia, three species were sampled (all belonging to G. sect. Gueldenstaedtia according to Tsui 1979), but these species were all treated to be G. verna (Georgi) Boriss. s.l. by some workers (Yakovlev 1988; Zhu 2004; Bao and Brach 2010). Further work is needed to test the delimitation of G. vernas.l.

Within Tibetia, two accessions of T. himalaica (Baker) H.P.Tsui grouped together, which were sister to T. yadongensis H.P.Tsui (Figs 1, 2). The tree topology and the morphological characters (e.g., elongate stem and round or retuse leaflets apex) seem to be consistent with treating T. himalaica as a distinct species (also see Tsui 1979; Cui 1998; Zhu 2005a; Bao and Brach 2010).