Taxon Sampling. We sampled 70 Meconopsis accessions for this study that represent every proposed section and series (Taylor, 1934) in the genus. Nine outgroup species (accessions) were selected and sampled based on previous phylogenetic studies of Meconopsis (Yuan, 2002) and Papaver (Carolan & al., 2006). Samples were collected from the wild, from the living collection in the Royal Botanical Garden at Edinburgh, and (with permission) from specimens in various herbaria. Species names, authorities, collection information (including localities), and sequence information are listed in Appendix 1. In addition, we included 19 Meconopsis accessions from Yuan’s (2002) study, and downloaded their trnL-trnF spacer and nrITS sequences from GenBank. Their vouchers and sequence information are also listed in Appendix 1. Genetic markers for our accessions that could not be successfully amplified or for Yuan’s (2002) accessions that were not available in GenBank were coded as missing data (Appendix 1).

DNA Extraction, PCR and Sequencing. Genomic DNA was extracted from silica-dried leaf materials or herbarium specimens using the DNeasy Plant Minikit (Qiagen, Valencia, California, USA). We chose nrITS and the cpDNA marker trnL-trnF spacer that had been shown to be phylogenetically informative in previous studies of Papaveroideae (Yuan, 2002; Carolan & al., 2006). We also selected the cpDNA marker rbcL because it is commonly used for molecular dating in basal eudicot families (Wikström & al., 2001; Anderson & al., 2005; Bell & al., 2010) and also showed sequence variations in the selected Meconopsis species we tested. Additionally, the cpDNA markers matK and ndhF were tested and selected because they were easy to amplify and significantly contributed to the resolution of the relationships at the sectional level in Meconopsis. PCR amplification was carried out in 12 μL reaction volume with 1–20 ng DNA, 1.0 unit of Taq polymerase (labmade, The University of Texas at Austin), 0.5X Failsafe Buffer B (Epicentre Biotechnologies, Madison, WI, USA), and 2.0 μmol/L primers. Forty-five PCR cycles were performed at 95° C for 30 seconds, 50° C for 45 seconds, and 72° C for 45 seconds for each cycle. Internal primers were designed for amplifying herbarium samples. All the primer pairs used are listed in Appendix 2. All of the PCR products were visualized on agarose gel containing Syber Safe DNA gel stain (Invitrogen, Eugene, Oregon, USA). Successfully amplified products were cleaned using ExoSap (Exonuclease I: New England Biolabs Beverly, MA, USA; Shrimp Alkaline Phosphatase: Progema, Madison, WI, USA) following the manufacturers’ protocols. Cleaned PCR products were sequenced using an ABI 3730 DNA Analyzer at the Institute for Cell and Molecular Biology Core Facility at The University of Texas at Austin. Amplifying primers were used for sequencing. In addition, internal primers were also used for sequencing if the amplicon was greater than 900 base pairs (i.e., rbcL, matK and ndhF).

Phylogenetic Analysis. Sequences were assembled in Geneious 5.5 (Biomatters, New Zealand), and aligned by Geneious Alignment with the default setting and 5 refinement iterations. Alignments were then reviewed and refined manually. The partition heterogeneity test (ILD test, Farris & al., 1994, 1995) was used to test pairwise combinability for each pair of chloroplast markers and for the combined chloroplast sequences versus nrITS. The ILD test was implemented in PAUP* version 4.0b10 (Swofford, 2002) with the setting of simple taxon addition, TBR branch swapping, and 1000 heuristic searches of the datasets that only included variable sites to generate a null distribution. The results of the ILD test suggested that the four cpDNA markers (rbcL, matK, ndhF and trnL-trnF spacer) are combinable (a p > 0.32 was found for the test of each pair), but the combined cpDNA dataset was not combinable with the nrITS dataset (p < 0.01). Therefore, we concatenated the four cpDNA markers.

Bayesian analyses were conducted for the nrITS and concatenated cpDNA data sets using MrBayes v3.1.2 (Huelsenbeck & Ronquist, 2005). Partition analysis was conducted only for the combined cpDNA dataset with each cpDNA marker treated as a separate partition. The evolutionary models of nucleotide substitution were first selected by jModelTest (Posada, 2008) under the Akaike information criterion (AIC), and we applied the models available in MrBayes v3.1.2 and most similar to the best fit models estimated by jModelTest for each gene partition: GTR+I for nrITS, GTR+G for rbcL, GTR+I+G for ndhF, GTR+G for matK, and GTR+G for trnL-trnF. Prior probability distributions on all parameters were set to the defaults. Twenty million generations were run using a Markov chain Monte Carlo (MCMC) method with four chains. Trees were collected every 100th generation. With 25% burn-in, both 50% and 80% majority-rule consensus trees were estimated to generate a posterior probability (pp) for each node.

Tree Incongruence Testing. We tested the topological conflict between the nrITS and cpDNA trees in order to evaluate the probability that the observed discordance was due to stochastic errors (for instance, data sampling and tree estimating errors) rather than to different evolutionary histories. The testing of tree incongruence is a well-studied field with several established techniques that can be used to explain disagreement between tree topologies (Planet, 2006). Many of these tests are philosophically as well as algorithmically different from one another, allowing researchers to choose one or multiple tests depending on the research purpose and the knowledge of the accuracy of the test. In this study, we were interested in detecting potential reticulate evolution by localizing significant disagreements between the cpDNA and nrITS phylogenies. We applied the approximately unbiased (AU) method (Efron, 1985; Efron & Tibshirani, 1998; Shimodaira & Hasegawa, 2001) to test competing hypotheses of different tree topologies. We selected this method for its type I error controls (Shimodaira, 2002). The AU test is a non-parametric method that calculates p-values for the candidate trees or hypotheses. A null distribution is generated based on bootstrap replicates of log-likelihoods with different replicate sizes.

In the AU test, the concatenated cpDNA dataset and nrITS datasets were treated as two independent partitions. Topologies (or nodes) subjected to testing were inferred from the tree outputs of the Bayesian analyses. PAUP* version 4.0b10 was used to generate the site likelihood scores of the unconstrained and constrained trees for the comparison. Two series of tests were completed: 1) an unconstrained nrITS tree was compared to a set of constrained nrITS trees, each of which was constrained by each of the recovered nodes found on the cpDNA consensus tree; 2) an unconstrained cpDNA tree was compared to a set of constrained cpDNA trees, each of which was constrained at each of the nodes on the nrITS consensus tree. The PAUP files of site likelihood scores for each tree were then reformatted for use in the program CONSEL (Shimodaira & Hasegawa, 2001) to perform the AU tests.

Phylogenetic Analysis. We obtained and analyzed 716 bp (432 variable) of nrITS, 1756 bp (358 variable) of matK, 1648 bp (231 variable) of ndhF, 1085 bp (588 variable) of trnL-trnF, and 1395 bp (55 variable) of rbcL sequences. The phylogenies are illustrated in Fig. 1, in which we show only the most closely related outgroup species, Papaver alpinum. The specimens used in Fig. 1 were assigned to species by W. Xiao. The recovered nrITS tree (Fig. 1B) shows an unresolved basal polytomy that is well resolved in the cpDNA tree (Fig. 1A). We found that species or clades on the cpDNA tree (Fig. 1A) were frequently located at discordant positions in the nrITS tree (Fig. 1B). For example, Meconopsis napaulensis (circled in Fig. 1) is a sister taxon to Meconopsis autumnalis (pp 1.00) on the cpDNA tree; but most closely related to Meconopsis ganeshensis (pp 1.00) on the nrITS tree.

Fig. 1.

cpDNA and nrITS Bayesian phylogenies of Meconopsis and tree incongruence test results. Names used for accessions were determined by W. Xiao and can differ from annotations of previous authors. All nodes (one by one) in both nrITS and cpDNA Bayesian trees were subjected to AU tests. Each red dot indicates a clade in which the monophyly of all its descendants disagrees significantly with the recovered nrITS tree topology (i.e., the unconstained nrITS tree had a significantly higher likelihood score than the nrITS trees constrained to monophyly of taxa derived from a red node); and each blue dot indicates a clade, in which, the monophyly of all its descendants disagrees significantly with the recovered cpDNA tree topology (the unconstained cpDNA tree had a significantly higher likelihood score than the cpDNA trees constrained to monophyly of taxa derived from a blue node).

Our cpDNA phylogeny (Fig. 1A), for the first time, resolved the relationships between different sub-groups of Meconopsis. This recovered phylogenetic structure is not consistent with any previously published Meconopsis infrageneric classifications. For discussion, we divided the cpDNA tree into five clades (Clades 1–4 and Group H) (Fig. 1A). The Clades 1–4 somewhat correspond to known chromosome numbers (Ratter, 1968; Ying & al., 2006; Kumar & al., 2013). In Clade 1, the only known chromosome number (M. bella) is 2n = 22. In Clade 2, the most frequent chromosome number is 2n =  84 with others varying from 2n = 74 up to 120. In Clade 3, the chromosome number is commonly 2n = 56, rarely 2n = 28. In Clade 4, the chromosome number is normally 2n = 56 or, rarely, 2n = 14. Each clade also represents a section in our new infrageneric revision for the genus (Xiao, 2013). Group H (Fig. 1) contains all the M. horridula (sensu Taylor 1934) accessions we sampled in this study. Because Grey-Wilson (2000, 2014) and other authors (Ohba & al., 2009; Yoshida & Boufford, 2010; Yoshida & al., 2011) favored subdivisions of Taylor’s M. horridula (1934), we show this clade in detail using the species names of Grey-Wilson (2000) in Fig. 2B. A few accessions could not be identified with certainty using Grey-Wilson’s (2000) concepts (our determinations are given in Fig. 1) because some of his supposedly key characters overlap between different “species.” The phylogenetic structures (Fig. 2) show that Grey-Wilson’s (2000) M. horridula is not monophyletic in either the cpDNA or nrITS tree (Fig. 2A, D). Our samples No.1, 2, 3, 8, 9, 10 and 11 (Fig. 2 A), determined as M. horridula sensu Grey-Wilson (2014), do not represent a monophyletic taxon on either cpDNA or ITS tree. However, all of the accessions in Fig. 2 which correspond to Taylor’s (1934) broad circumscription of M. horridula form a monophyletic group in both the cpDNA and nrITS trees (Fig. 1, Group H).

Fig. 2.

Phylogenies of the Meconopsis horridula complex. Top figures: (A) Tree topology of the M. horridula complex from the Bayesian analysis of the cpDNA markers. (B) The taxa included are given names as they would be applied following Grey-Wilson’s (2000) treatment and arranged according to their position in the nrITS tree shown in Column D. A “?” indicates that the specimen could not be determined with certainty using the descriptions and geographic ranges provided by Grey-Wilson (2000). [Note: Our proposed names for these accessions are given in Fig. 1, Group H]. The dashed lines in Column A connect the terminal taxa on the cpDNA tree to their accessions in Column B. The symbols in Column A are used to plot the accessions on the map (bottom). Specimens marked by the same symbol come from the same clade in the cpDNA tree. (C) Elevations of the accessions in Column B. (D) Tree topology of nrITS Bayesian analysis. The clades in this tree are color coded and mapped in the bottom figure. The map also shows the location of each accession using their associated numbers in Column A.

We illustrate the main taxonomic discrepancies between Taylor (1934) and Grey-Wilson’s (2006) treatment of Meconopsis napaulensis in Fig. 3. The phylogenetic relationships among the included species (or accessions) are inferred from Fig. 1. According to either the cpDNA or nrITS phylogeny, Taylor’s (1934) M. napaulensis does not reflect a monophyletic group while Grey-Wilson’s (2006, 2014) treatment is consistent with our phylogenetic results.

Fig. 3.

The superimposed cpDNA (green) and nrITS (black) trees showing discrepancies in the placements of Meconopsis napaulensis and its allies. The right two columns show the probable identifications of the accessions using either the key provided by Taylor (1934) or following Grey-Wilson (2014). The red arrows point to putative hybridization events.

Tree Incongruence Testing. Our results testing if the disagreements between the cpDNA and nrITS tree were statistically significant (indicated by the colored dots on the Bayesian trees in Fig. 1), showed that when constraining to monophyly all the taxa derived from a red node (labeled on the cpDNA tree, Fig. 1A), the constrained nrITS tree had a significant lower likelihood score (p < 0.01) than the unconstrained nrITS tree; and similarly, by constraining all the taxa derived from a blue node on the nrITS tree (Fig. 1B), a significant difference of likelihood score resulted between the unconstrained and constrained cpDNA trees. These results localized the taxa that caused the significant incongruence between the cpDNA and nrITS trees.

Meconopsis horridula Complex. In the first monograph of Meconopsis, Taylor (1934) treated Meconopsis horridula as a polymorphic species by aggregating seven previously described species, among which Meconopsis racemosa, Meconopsis rudis, Meconopsis prattii, and Meconopsis horridula had been the most widely recognized taxa. Taylor’s (1934) treatment was based on his observations that a wide range of intermediate forms bridge the extreme forms across his concept of M. horridula, with no satisfactory distinctions. In light of our phylogenetic results, Taylor’s treatment (1934) cannot be rejected because all of the accessions of what he recognized as M. horridula were monophyletic in both the cpDNA and nrITS trees (Group H, Fig. 1).

Grey-Wilson (2000) divided the Taylor’s horridula into two widely distributed species, M. horridula and M. prattii, and a narrowly endemic species M. rudis. In his treatment, M. horridula was described as a high altitude species of short stature (< 40 cm) with predominantly scapose flowers and M. prattii as a tall species (30–70 cm) growing at relatively low elevations with all flowers arising along a central stem. Grey-Wilson (2000) diagnosed M. rudis by its bluish green leaf blades and a unique dark purple color at the bases of the leaf spines. Defined as such, M. rudis occurs only in northwestern Yunnan Province (China) centered on the Yulong Mountain. We found that neither the cpDNA nor nrITS phylogeny supports (Fig. 2) the monophyly of M. horridula sensu Grey-Wilson (2000). In addition, closely related specimens within the same clade (in both the cpDNA or nrITS trees, Fig. 2A, D) are frequently found across a wide elevational range, indicating that there is no clear altitudinal boundary to subdivide Taylor’s broad concept of M. horridula (1934) in contrast to Grey-Wilson’s (2000) assertion.

Grey-Wilson (2014) included eleven species of Meconopsis referable to Taylor’s (1934) M. horridula, and organized them into two series under his section Racemosae: series Heterandrae includes only Meconopsis heterandra and Meconopsis balangensis; and series Racemosae includes the rest of the section. In Fig. 2B, it is shown that M. heterandra and M. balangensis are not related. Thus, neither of the series Heterandrae nor series Racemosae is monophyletic. Our samples 1, 2, 3, 8, 9, 10 and 11 (Fig. 2) were determined as M. horridula sensu Grey-Wilson (2014) which appears as a polyphyletic group on either the cpDNA or ITS tree.

One plausible reason for the difficulty in dividing Taylor’s M. horridula is revealed in Fig. 2. There is frequent incongruence between the two trees for these accessions, suggesting a history of frequent reticulation within this widely distributed group of taxa. Such a pattern of gene flow could explain Taylor’s recognition of a broadly defined M. horridula in which he stated that it is difficult to assign a large range of intergrading morphologies into meaningful subdivisions. We, therefore, propose that Taylor’s (1934) concept of M. horridula is best considered as a species complex which can serve as a guide for future phylogeographic studies: this species complex with its wide geographic distribution, remarkable ecological plasticity and morphological diversity (Taylor, 1934) could be a model for understanding the evolutionary and biogeographic history of many plant taxa that extend across the Himalaya through the Tibetan Plateau to the Hengduan Mountains. This M. horridula complex would then include the following named taxa:

Meconopsis napaulensis. A series of taxonomic conflicts has also centered around Meconopsis napaulensis. Unlike the situation in the Meconopsis horridula complex, the problems in M. napaulensis are not simply a matter of how to establish species boundaries but rather different interpretations of the identity of M. napaulensis. The species was described by Augustin Pyramus de Candolle in 1824, however, the type specimen is fragmentary and has no flowers. The lack of important characteristics and detailed information on the label, as well as a cursory description, have led to different opinions about assigning specimens to M. napaulensis. Between the two opinions of M. napaulensis from Taylor (1934) and Grey-Wilson (2006, 2014), Grey-Wilson (2006) justified his interpretation of M. napaulensis by arguing that he matched the locality, the morphology of the fruit, and indumentum of the type of M. napaulensis to some specimens collected from Ganesh Himal (a sub-range of the Himalaya in north-central Nepal). Taylor (1934), in contrast, presented little detail of his reason for assigning specimens to M. napaulensis. Thus, Grey-Wilson’s (2006) work provides a more convincing interpretation of the circumscription and identity of M. napaulensis than Taylor’s (1934).

We selected a few specimens representing both Taylor’s (1934) and Grey-Wilson’s (2006, or 2014) species concepts of Meconopsis napaulensis. The phylogenetic relationships of these specimens and how they would have been identified by Taylor and Grey-Wilson are shown in Fig. 3. We can see that Taylor’s M. napaulensis is split with specimens re-assigned to M. staintonii, M. wilsonii and M. wallichii by Grey-Wilson (2006, or 2014). The latter two, accommodating the majority of the specimens Taylor annotated as M. napaulensis, are not related to M. staintonii (Fig. 3), which indicates that Taylor’s (1934) M. napaulensis does not represent a monophyletic taxon.

Taylor (1934) had access to only a limited number of specimens, but he placed plants with “non-yellow (e.g., red, blue) petals” into two species: Meconopsis napaulensis (a widely distributed species) and M. violacea (a species of restricted distribution). Taylor’s classification cannot accommodate some recent collections (collected after 1935), as indicated in Fig. 3, but it had a lasting impact: most (if not all) of the red-flowered plants (e.g., M. staintonii, M. ganeshensis, M. wallichii, and M. chankheliensis) were originally collected and first determined as M. napaulensis, and/or introduced to British gardens as such. These plants, however, are not members of the same clade (Fig. 1). Grey-Wilson’s treatment (2006, or 2014) corrected the classification by assigning these red-flowered plants to different taxa. Given that Grey-Wilson’s taxonomy of M. napaulensis and its allies (2006) is compatible with our phylogenetic results, we have followed it in our work.

For future studies, more precise delimitations between Meconopsis napaulensis and its allies is desirable, for even Grey-Wilson’s latest treatment (2014) does not clearly address the potential of gene flow between the species he recognizes. For example, Meconopsis regia, Meconopsis paniculata, and Meconopsis staintonii are known to hybridize freely with each other in the garden (Grey-Wilson, 2006). Taylor (1934) treated Meconopsis wallichii (sensu Grey-Wilson 2006) and Meconopsis wilsonii (sensu Grey-Wilson 2006) as a single species (i.e., his M. napaulensis) and pointed out a flux of continuous forms that cannot be satisfactorily assigned to species. Moreover, the placement of M. napaulensis (i.e., accession X078) caused significant disagreement between the nrITS and cpDNA trees (Fig. 1), and this M. napaulensis accession grows in the same area as M. autumnalis and M. ganeshensis. This pattern of nuclear and organelle tree incongruence indicates a need for further study to resolve whether each of the three named taxa is a distinct lineage.