1. Introduction

The genus Cymbidium SW., with approximately 80 species, is primarily distributed throughout the subtropics and tropical areas of Asia and northern Australia (Du Puy and Cribb, 2007; Chen et al., 2009; Pridgeon et al., 2009). In China, more than 50 species are found according to the most recent revision and recently published new species (Liu and Chen, 1998, 2002, 2004, 2005; Long et al., 2003; Liu et al., 2006; Chen et al., 2009; Lan et al., 2018; Zhang et al., 2018). Cymbidium is one of the earliest orchid groups to be cultivated, making excellent potted plants and cut flowers due to its extremely high ornamental and economic value. Commercially important hybrids have been cultivated for a long time in China and adjacent regions (Liu et al., 2006). Despite its acknowledged importance, limited and often ambiguous morphological differences pose challenges to understanding intergeneric relationships within Cymbidium. The intergeneric relationships of Cymbidium remain an open question due to limited and often ambiguous morphological differences.

Since the establishment of Cymbidium by Swartz in 1799, various generic delimitations and infrageneric systems have been proposed based on morphological characters. Dressler (1981) places Cymbidium in Cymbidieae of Vandioideae, which contains all of the sympodial vandoid orchids, mostly with two pollinia. Schlechter (1924) proposes an infrageneric system of Cymbidium with eight sections, which is the basis of the modern infrageneric classification of Cymbidium, and most sections are still being recognized more or less in their original form. Hunt (1970) included Cyperorchis within Cymbidium and maintained Schlechter's sectional divisions. Seth and Cribb (1984) initially divided Cymbidium into three subgenera based on the number of pollinia and the state of fusion between lip and column: subgenus Cymbidium with two pollinia and free lip, subgenusCyperorchis with two pollinia and fusion of the lip and column base, and subgenus Jensoa with four pollinia and free lip. Puy and Cribb (1988) slightly modified this treatment and added section Borneensia for the recently described Cymbidium borneense Wood. Liu et al. (2006) followed the treatment of Puy and Cribb (1988) with some modifications and additions, added sections Nanula and Axillaria, transferred the section Borneensia from subgenus Cymbidium into the subgenus Jensoa, and reduced the section Maxillarianthe to synonymy of the section Jensoa.

Recently, the results of molecular analyses have shed new light on the taxonomy of Cymbidium (Cameron et al., 1999; Berg, 2002; Yukawa and Stern, 2002; Zhang et al., 2002; Sharma et al., 2012; Yang et al., 2013; Lan et al., 2018). The phylogeny of Cymbidium reconstructed by Zhang et al. (2002) based on the analyses of nrITS sequences in 30 taxa belonging to three subgenera showed that the genus was split into several clades and intermixed with the main subgenera, suggesting that the previous division among three subgenera should be evaluated with more data. Berg (2002) performed a molecular phylogenetic analysis of Cymbidium using one cpDNA marker (matK) and nrITS and indicated that two or three subgenera can potentially be defined within the genus and affirmed a southeast Asian origin for the genus.Yukawa and Stern. (2002) obtained the same result in their strict consensus tree based on nrITS and matK sequences. Assessment of phylogenetic interrelationships in the genus Cymbidium from northeast India based on nrITS showed that this genus was divided into three subgenera (Sharma et al., 2012). Du Puy and Cribb (2007) according to the DNA studies by Berg (2002) and Yukawa and Stern (2002), concluded that the subgenera of Cymbidium were not monophyletic, and they retained the sections but dispensed with subgenera.

All of the pre-DNA era classifications of Cymbidium were based on a relatively small set of morphological aspects and features, especially on the lip and pollinium numbers, which have led to considerable taxonomic uncertainty and debates (Dressler, 1993; Freudenstein and Rasmussen, 1999). As previous results of molecular systematics were largely based on sparse sampling across Cymbidium or mainly utilized either a single DNA marker (especially nrITS) or two markers (nrITS and matK), some conclusions and results were weakly supported or even without statistical support (Cameron et al., 1999; Yukawa and Stern, 2002; Zhang et al., 2002; Yang et al., 2013; Lan et al., 2018). Therefore, it is necessary to understand the relationships within Cymbidium and the delimitation of the infrageneric taxa to base the analyses on multiple DNA markers and a denser sampling acrossCymbidium.

One additional challenge to accurately reconstructing phylogenetic relationships is the possibility of phylogenetic incongruence between cpDNA and nuclear sequence data (Tu et al., 2008; Pelser et al., 2010; Guo et al., 2015; Tang et al., 2015). Phylogenetic incongruence may be a result of stochastic errors, systematic errors, incomplete lineage sorting (ILS), introgressive hybridization (IH), paralogous gene sampling, or horizontal gene transfer (HGT) Francine et al., 2017, Geuten et al., 2004, Richardson and Palmer, 2007, Russell et al., 2010, Yang et al., 2012.

In this study, we reconstructed the phylogeny of 70 representatives in three subgenera of Cymbidium using both plastid DNA (rbcL, trnS, trnG, matK, trnL, psbA, and atpI) and nrITS sequences. Our goals were to (1) establish a phylogeny based on the seven plastid DNA and one nuclear gene for the Cymbidium with significantly increased taxa sampling with an emphasis on Chinese species, (2) identify relationships that were inconsistent between nuclear and plastid trees, and (3) explore possible causes of the incongruence.

2. Material and methods 2.1. Taxon sampling

To assess phylogenetic relationships within Cymbidium, we used seven plastids (rbcL, trnS, trnG, matK, trnL, psbA and atpI) and one nuclear (nrITS) marker sampled from 70 species (varieties) of Cymbidium. Four species from the tribe Malaxideae Lindley (Lindley, 1826) and one species from the tribe Vandeae Lindley (Lindley, 1821) were used as outgroups. The selection of outgroups was based on the classification of Chase et al. (2015). In this study, 510 sequences (468 of which were newly sequenced) were obtained, the voucher information and GenBank accession numbers were listed in Table 1, and the specimen was deposited in the herbarium of the National Orchid Conservation Center of China (NOCC).

2.2. Collection of DNA sequences

Total DNA was extracted from fresh material using a modified CTAB procedure of Doyle and Doyle (1987). DNA extraction, PCR amplification, and sequencing were performed according to Chen et al. (2017). The primers used for PCR analysis were listed in Table S1.

2.3. Sequence analysis and alignment

Both forward and reverse sequences referring to the corresponding chromatograms were edited and assembled into contig sequences using SeqMan v.7.1 (DNAStar, USA) with the default "Classic Assembler" parameters (Match Size = 12; Minimum Match Percentage = 80). DNA sequences were aligned with MEGA 5.05 under the Muscle model and manually adjusted to account for obvious or missing inserts (Tamura et al., 2011; Zhang et al., 2013).

2.4. Identification of incongruence

The congruence among the nuclear data (nrITS) and the combined chloroplast DNA data set (rbcL, trnS, trnG, matK, trnL, psbA and atpI) was tested using the incongruence length difference (ILD) test (Farris et al., 1995), implemented as the Partition Homogeneity test in PAUP* v.4.0b10, and followed procedures described by Li et al. (2015). Incongruence was also visually inspected for in trees that exhibited contrasting topologies when obtained from different data sets. The thresholds of hard incongruence followed those adopted by Pelser et al. (2010): bootstrap values ≥ 80 and/or PP ≥ 0.95, as well as ILD P < 0.01.

2.5. Phylogenetic analyses

Phylogenetic analyses were performed using Bayesian inference (BI) and maximum-likelihood (ML) methods. The evolutionary models for the ML and BI analyses were determined by jModelTest using the Akaike Information Criterion (AIC). ML analysis was performed using the CIPRES Science Gateway web server (RAxML-HPC2 on XSEDE 8.2.10) (Miller et al., 2010) with 1000 bootstrap replicates and settings that are described inEdgar. (2004). BI analysis was performed using the CIPRES Science Gateway web server (MrBayes 3.2.6 on XSEDE) (Stamatakis et al., 2008). The following settings were used: sampling frequency = 1000; tem = 0.1; burn-in = 2000; and number of Markov chain Monte Carlo generations = 10, 000, 000 (Li et al., 2015).

2.6. Network analyses

To visualize conflicts among gene trees, SPLITSTREE4 v.4 (Huson and Bryant, 2006) was used to generate a consensus tree (seven cpDNAs and one nuclear gene).

3. Results 3.1. Sequences and alignment

In the present study, 468 new Cymbidium sequences were obtained from 15 sections (except section Borneensia) of three subgenera, the division of sections within Cymbidium was adopted according to Liu et al. (2006). However, we were unable to amplify the chloroplast regions of a few accessions; we treated these as missing data. Aligned sequence length were as follows: 700 bp for the nrITS region (111 bp parsimony-uninformative and 189 bp parsimony-informative in the data set), 1293 bp for rbcL, 1626 bp for matK, 768 bp for trnG, 1027 bp for trnS, 1314 bp for trnL, 955 bp for psbA, and 838 bp for atpI. A total of 869 bp of the combined plastid regions (7524 bp) were parsimony-uninformative; 1004 bp were parsimony-informative. The numbers of variable and parsimony informative sites were listed in Table 2. Details pertinent to the best-fit model of molecular evolution could be found in Table 3.

3.2. Phylogenetic analysis of the combined cpDNA data set

Our phylogenetic analyses of cpDNA focused on the combined data set. Seventy-one taxa were included in this cpDNA matrix, five of which were outgroups. Four major groupings within Cymbidium were recovered with moderate support (Fig. 1). Clade A consisted of one species of subgenus Cymbidium, three species of subgenus Jensoa and 23 species of subgenus Cyperorchis. Clade B was composed of one species of subgenus Cymbidium and one species of subgenus Jensoa, Cymbidium dayanum Rchb. f. and Cymbidium omeiense Y.S. Wu et S.C. Chen. Clade C was composed of nine species, seven species of subgenus Cymbidium, and two species of subgenus Cyperorchis (PP = 0.91). Clade D contained three species of subgenus Cymbidium, one species of subgenus Cyperorchis, and 24 species of subgenus Jensoa (PP = 0.91).

3.3. Phylogenetic analysis of the nrITS data set

Seventy-five samples, including five species identified as outgroups, were sampled for the nrITS analysis, and three differentiation clades (clades A-C) were recovered (Fig. 2).

(1) Clade A included nine species from two sections of subgenus Cymbidium clustered in a single lineage (PP = 0.9, BS = 68): C. rectum Ridley., C. mannii Richb. f., C. paucifolium Z.J. Liu et S.C. Chen, C. bicolor Lindl., C. canaliculatum R.Br., and C. aloifolium (L.) Sw. C. finlaysonianum Lindl., C. atropurpureum (Lindl.) Rolfe, and C. puerense Z.J. Liu et S.R. Lan.

(2) Clade B consisted of 27 species from two subgenera and was poorly resolved with low support (PP = 0.55, BS = 52). It was subdivided into eight subclades (subclades 1–8). Subclade 1, which was sister to the other seven subclades, and included two subgenera: Cymbidium and Cyperorchis, with weak support. Subclade 2 contained eight species from five sections: Iridorchis, Eburnea, Annamaea, Cyperorchis, and Parishiella with weak support. The subclade 3 consisted of two species of section Cyperorchis: Cymbidium wenshanense and C. quinquelobum, with strong support (PP = 1, BS = 100). Subclade 4 only included one species of section Himantophyllum from subgenus Cymbidium, C. dayanum, with strong support (PP = 0.91). The subclade 5 (PP = 0.91, BS = 66) was composed of four species from section Iridorchis. Subclade 6 included two species from section Iridorchis. Subclade 7 was composed of two species of section Cyperorchis and one species of section Iridorchis and was sister to subclade 8, which consisted of four species of section Eburnea; however, the relationships had low support.

(3) Clade C (PP = 0.52, BS = 68) was composed of 27 species and seven un-identified species from two subgenera, subgenus Cymbidium and subgenus Jensoa. Six divergent subclades (subclades 9–14) were recovered in this clade. Subclade 9 included six species of subgenus Cymbidium, with strong support (PP = 1, BS = 92). The subclade 10 was composed of two species of section Geocymbidium and one species of section Pachyrhizon. Subclade 11 only included one species of section Pachyrhizon. Subclade 12 was composed of nine species from three sections (sections Axillaria, Jensoa, and Nanula) with strong support (PP = 1, BS = 97), but the interrelationships were poorly resolved with low support. Subclade 13 consisted of Cymbidium cyperifolium Wall. et Lindl., C. defoliatum Y.S. Wu et S.C. Chen, and C. faberi Rolfe, and one unidentified species was sister to the remaining members of this clade. Subclade 14 consisted of 11 species from two subgenera (Jensoa and Cymbidium), but the interrelationships were poorly resolved with low support.

3.4. Incongruence tests

Our molecular analyses using nrITS and a combined cpDNA data sets indicated many topological conflicts, some of which appear to be quite strong judging by support values. The phylogenetic tree based on the nrITS data was divided into three clades, but it could be divided into four clades based on the seven cpDNA sequences. The nuclear gene trees indicated that subgenusCymbidium diverged first, followed by subgenus Cyperorchises with two species of subgenus Cymbidium and finally the subgenus Jensoa with two species of subgenus Cymbidium and one species of subgenus Cyperorchis. However, the cpDNA gene trees indicated that subgenus Cyperorchis diverged first with one species of subgenus Cymbidium and three species of subgenus Jensoa, followed by the clade C. dayanum-C. omeiense, followed by the subgenus Cymbidium with two species of subgenus Cyperorchis, and finally the subgenus Jensoa with three species of subgenus Cymbidium and one species of subgenus Cyperorchis.

Moreover, there were many topological conflicts within sections. For example, the section Floribunda was monophyletic in the nrITS gene tree, with strong support (PP = 0.99, BS = 97). This section was subdivided into two clades in the cpDNA tree. The section Cymbidium was monophyletic in the cpDNA tree and was sister with Cymbidium canaliculatum. However, in the nrITS gene tree, C. canaliculatum nested within this section, creating a polyphyletic group. There were also many topological conflicts in species, such as Cymbidium faberi, Cymbidium aestivum, and Cymbidium daweishanense. The ILD test for the nrDNA and combined cpDNA data resulted in P < 0.01 and indicated incongruence between the two data sets; therefore, we did not concatenate these two data sets.

3.5. Network analysis

The generated networks revealed inter- and intrasectional reticulations in Cymbidium (Fig. 4). The three subgenera (Cymbidium, Cyperorchis, and Jensoa) that exhibited discordant phylogenetic positions in the separate gene trees formed complex networks, suggesting that hybridization events occurred between these three subgenera (Fig. 4a). When some species with highly diverged alleles were excluded, the network became much simpler, but reticulation was still observed (Fig. 4b).

4. Discussion 4.1. The phylogeny of the Cymbidium

In the present study, an updated phylogeny of Cymbidium was proposed based on comprehensive sampling of 75 species (varieties) (five of which were outgroup) and 468 new DNA sequences. The overall results of the phylogenetic analysis of Cymbidium were consistent with previously published results (Berg, 2002, Yang et al., 2012, Yukawa and Stern, 2002). Based on the nuclear DNA (nrITS) data set, Cymbidium was found to be composed of three major clades, while it was divided into four clades in the cpDNA tree.

We investigated the relationships among the subgenera of Cymbidium. The monophyly of subgenus Cymbidium was broken, divided into five clades, four of which were nested in two other subgenera (Cyperorchis and Jensoa). The monophyly of subgenus Jensoa was broken by the nested position of subgenus Cymbidium section Himantophyllum, C. aestivum, and subgenus Cyperorchis, C. daweishanense. The principal synapomorphic character of the subgenus Jensoa is having four pollinia (Du Puy and Cribb, 2007, Liu et al., 2006); however, C. aestivum and C. daweishanense do not share this character. If it were not for the positions of Cymbidium devonianum and C. dayanum, subgenus Cyperorchis would be monophyletic. Transfer of C. devonianum and C. dayanum to a member of subgenus Cyperorchis resulted in the loss of a prominent synapomorphic character of subgenus Cyperorchis, that is, a fused basal part of the lip with the column. Taking the morphological characteristics and phylogenetic relationships into account, subdivisions of Cymbidium at the subgeneric level were not useful. We did not find any stable synapomorphic characters that are featured in each subgenus clarified in this study. The delimitation of sections within Cymbidium were also problematic, and most sections were found to be polyphyletic. In summary, the currently defined subgenera and sections of Cymbidium are not monophyletic. Furthermore, the phylogenetic relationships among Cymbidium have not been resolved. Additional markers or second- or third-generation sequencing may be required for further study.

4.2. Reticulate evolution in Cymbidium

The evolutionary relationships at the species level and above could be idealistically represented with bifurcating phylogenetic trees based on the theory of universal common descent. Some evolutionary events, such as horizontal gene transfer, lineage sorting, rapid radiation, hybridization and introgression, may result in conflicts among gene trees, and therefore, phylogenetic networks are more suitable to model the real relationships among species (Geuten et al., 2004; Russell et al., 2010; Yang et al., 2012; Guo et al., 2015).

Phylogenies of Cymbidium obtained from nrITS and cpDNA sequences were concordant in certain respects and discordant in others (Fig. 3). Both phylogenies support non-monophyly of each of the three subgenera of Cymbidium, as well as sections. The generated networks revealed inter- and intrasectional reticulations in Cymbidium (Fig. 4). The three subgenera (Cymbidium, Cyperorchis, and Jensoa) that exhibited discordant phylogenetic positions in the separate gene trees formed complex networks, suggesting that hybridization events occurred between these three subgenera (Fig. 4a). When we deleted the 24 highly divergent alleles, the reticulation relationship among Cymbidium was still very complex, indicating that species within Cymbidium underwent complex reticulate evolution (Fig. 4b). Actually, due to the sympatric distribution, weak reproductive isolation of the species, and synchronous flowering, the natural interspecific hybridization within orchids occurs commonly (Cribb, 1998; Liu et al., 2009; Guo et al., 2015). Moreover, there have been 15 natural hybrid species reported in the Cymbidium (http://www.emonocot.org/). Furthermore, thousands of artificial interspecific hybrids are listed by the Royal Horticultural Society (http://apps.rhs.org.uk/horticulturaldatabase/orchidregister/orchidregister.asp). This phenomenon, which is caused by absence of strong interspecific reproductive barriers and hybrid zones, has also been found in other orchid genera, such as Orchis Tourn. ex L. (Bateman et al., 2008), Ophrys L. (Cortis et al., 2009), Epidendrum Pav. ex Lindl (Pinheiro et al., 2010), and Paphiopedilum Pfitzer (Guo et al., 2015). Accordingly, hybridization may play an important role in orchid speciation.

4.3. The possible causes of tree incongruence and the low value

When comparing the nuclear and combined plastid trees (Fig. 3), we immediately observed conflicting branches, but the most branches did not have strong support. The ILD test for the nrDNA and combined cpDNA data resulted in P < 0.01, indicating incongruence between the two data sets. In addition, to further determine incongruence of phylogeny tree, we generated the network, which showed that there were complex networks inCymbidium. Tree incongruence and low support values may be explained by hybridization inCymbidium and the lack of informative characters.

4.3.1. Hybridization

Hybridization has long been reported in plant lineages and appreciated to be a key mechanism in plant evolution, as many extant taxa have likely recently originated from hybridizations (Zhang et al., 2012; Guo et al., 2015; Francine et al., 2017). Sequences from different genomes of hybrid species usually reflect different lines of inheritance (e.g., mitochondrion genes from the paternal line, plastid genes from the maternal line, and nuclear genes from both parental lines), which could result in the incongruence between these different data sources (Yu et al., 2013; Zhai et al., 2014; Guo et al., 2015; Kanzi et al., 2020). In Cymbidium, we speculated that hybridization may be a cause of tree incongruence. Good examples are described taxa such as Cymbidium × nishiuchianum, Cymbidium × purpuratum, Cymbidium × latifolium, Cymbidium × uniflorum, Cymbidium × oblancifolium and Cymbidium × nujiangense. This phenomenon is also found in Calanthe (Zhai et al., 2014), Epidendrum (Pinheiro et al., 2010), and Paphiopedilum (Guo et al., 2015).

4.3.2. The lack of informative characters

Although we used eight genes to construct the phylogenetic relationships ofCymbidium, the informative characters available in these eight genes were limited, and they might represent only a small piece of the evolutionary story in Cymbidium. Moreover, the support values for most clades were low. The lack of informative characteristics may be one cause of topological incongruence between the plastid and nuclear phylogenies (Tang et al., 2015).

5. Conclusion

The present work clarified the phylogenetic relationships within Cymbidium through molecular evaluations. There were incongruent results in the topology of the combined chloroplast and nrITS trees, and the support values of clades were low. These findings may be the result of natural hybridization and a lack of informative characters. In addition, we detected reticulate evolution in Cymbidium. These results add valuable insights into the evolution of Cymbidium. Additional studies based on second- or third-generation sequencing are needed, with a focus on geographic and ecological patterns and the tempo and mode of evolution in the genus. This genus is mainly distributed in subtropical and tropical areas of Asia, and the investigation of diversification patterns of this genus will shed light on biodiversity evolution in this region.

Author contributions

GQZ and JWZ designed the experiments. JH and XYW performed the experiments. GQZ and GZC contributed to the data analysis and molecular system construction. LJC drew the figures. GZC wrote the manuscript. WHR contributed to the collection and morphological identification of some samples. ZJL, SRL, DHP, and MHL provided suggestions on the experimental design and discussion sections. All the authors read and approved the final manuscript.

Declaration of competing interest

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as potential conflicts of interest.

Acknowledgments

This research was funded by The National Key Research and Development Program of China (Nos. 2018YFD1000401 and 2018YFD1000400), the Natural Science Foundation of Guangdong Province (No. 2017A030312004), and the Science and Technology Plan Project of Shenzhen (No. JCYJ20170817151501595). The authors thank Xu-Hui Chen and Wei-Rong Liu for their support in data collection and the morphological identification of some samples and Fu-Ye Wen for the daily management of samples.

Appendix A. Supplementary data

Supplementary data to this article can be found online at https://doi.org/10.1016/j.pld.2021.08.002.