Reconstructing polyploid evolution and ancestral chromosome numbers

Phylogeny reconstruction – We firstly calculated a cpDNA tree to serve as the phylogenetic framework for ancestral chromosome number reconstruction. Because Meconopsis is closely related to Papaver and Roemeria (Xiao, 1913), we included Papaver and Roemeria species with Cathcartia as outgroups, in order to achieve an estimation of rates of dysploidy formation and/or genome duplication rates. This phylogenetic reconstruction employed the concatenated sequences (trnL-trnF, matK, ndhF and rbcL) of Meconopsis and Cathcartia species, obtained from our phylogenetic study of Meconopsis (Xiao, 2013) (sequence information in Appendix 1); and 27 previously published trnL-trnF sequences downloaded from Genbank (sequence information is listed in Appendix 2), mainly including Papaver and Roemeria species. Sequences were assembled in Geneious 5.5 (Biomatters, New Zealand), and aligned by Geneious Alignment with the default settings. Absent markers were treated as missing characters. Alignments were then refined manually. Bayesian partition analysis was applied using MrBayes 3.1.2 (Huelsenbeck & Ronquist, 2001), with each marker treated as a partition. The evolutionary models were selected using jModelTest (Posada, 2008). We applied the models most similar to the best-fit models estimated by jModelTest that were available in MrBayes v3.1.2 for each partition: GTR+G for rbcL, GTR+I+G for ndhF, GTR+G for matK, and GTR+I+G for trnL-trnF dataset. Prior probability distributions on all parameters were set to the defaults. Stationarity was reached at twenty million generations using a Markov chain Monte Carlo (MCMC) method. Trees were collected every 100th generation. With 25% burn in, a 50% majority-rule consensus tree was calculated.

Previous cytological studies (Ratter, 1968; Kadereit, 1987; Lavania & Srivastava, 1999; Ying et al., 2006; Kumar et al., 2013) have provided chromosome numbers for many species. For estimating ancestral characters, parsimony based methods have been widely adopted but parsimony approaches usually suffer from strong bias due to their inability to incorporate multiple models and to deal with uncertainty. Mayrose et al. (2010) developed a likelihood method implemented in chromEvol, which estimates the probability of a given ploidy level at any internal node in a given tree. In chromEvol, eight different models are available for testing, each representing a different hypothesis by estimating a different set of parameters (Table 1). For example, the model CONST_RATE_DEMI is characterized by parameters of a constant rate for gaining a single chromosome (gainCONSTR), a constant rate for losing a single chromosome (lossCONSTR), and a constant rate for both whole genome duplication (duplConstR) and demi-polyploidization (demiPloidyR). This model thus hypothesizes that dysploidy, whole genome duplication, and demi-polyploidization (change from 2n to 3n, e.g., forming a triploid) all have occurred along a phylogeny at a constant rate. In contrast to the CONST_RATE_DEMI model, some other models do not allow whole genome duplication or demi-polyploidization. Some of these estimate linear rates (instead of constant rates) to implement the hypothesis that chromosome number change rate is dependent on the current chromosome number. Four of these models were also used. Detailed model comparisons can be retrieved from  http://www.tau.ac.il/∼itaymay/cp/chromEvol/index.html. All the eight models were run to select for the best-fit model. The selected model was used to estimate ancestral chromosome numbers.

Table 1.

Model test result in chromEvol (sorted by AIC score).

Reconstructing a GAPDH phylogenetic network to detect ancient reticulation

Amplification, PCR, cloning and sequencing – We screened the EST library of Papaver somniferum and related species to choose genes with a single or low-copy number. Among the candidate genes, GAPDH (partial) was selected for its relatively low copy number (we obtained only one copy of the GAPDH gene for the accession of Papaver alpinum, a diploid taxon), successful PCR amplification, and the possession of four introns (five exons) that are phylogenetically informative. Genomic DNA was extracted from silica-dried leaf materials or herbarium materials using the DNeasy Plant Minikit (Qiagen, Valencia, California, USA). Primers were designed with the reference to the EST sequences of Papaver somiferum and other species in Papaveraceae using the EST database from GenBank. The primer pair, forward primer 5′-CACCACCAACTGTCTTGCTCCCCT-3′ and reverse primer 5′-AGCACCCACACTGAAGAGGGAC-3′, were selected. PCR amplification conditions were optimized and carried out in 25 µL reaction volumes with 20–40 ng DNA, 1.0 unit of Taq polymerase (made by the author), 0.5X Failsafe Buffer B (Epicentre Biotechnologies, Madison, WI, USA), and 2.0 µmol/L primers. Eighteen PCR cycles were performed at 95° C for 30 seconds, 59°C for 40 seconds, and 72° C for 45 seconds for each cycle; followed by 26 cycles of 95° C for 30 seconds, 55°C for 40 seconds, and 72° C for 45 seconds. PCR products were visualized on agarose gel containing Syber Safe DNA gel stain (Invitrogen, Eugene, Oregon, USA). Successfully amplified products were cloned using the TOPO-TA cloning kit (Invitrogen, Carlsbad, CA, USA). Thirty to 60 colonies were picked from each plate and amplified by M13 plasmid primers from the cloning kit with the manufacturers' protocols. Successfully amplified products were cleaned using ExoSap (Exonuclease I: New England Biolabs Beverly, MA, USA; Shrimp Alkaline Phosphatase: Progema, Madison, WI, USA) with the manufacturers' protocols. Cleaned PCR products were sequenced on an ABI 3730 DNA Analyzer at the Institute for Cell and Molecular Biology Core Facility at The University of Texas at Austin. GAPDH sequence information is listed in Appendix 3.

Test for recombinant sequences – A potential difficulty of using nuclear markers is that if Meconopsis were a highly polyploid group, multiple copies of the target gene would be amplified simultaneously. Under these circumstances, artificial recombinants could be introduced during PCR when multiple gene copies coexist. We applied an assumption when screening out PCR-mediated recombination that recombinants can only occur once in the raw sequence data but sequences of true genes can reoccur. This assumption is based on the mechanism of PCR-mediated recombination that interruption of complete extension (for example, Taq polymerase can stop extension when mismatching occurs) during PCR produces an incomplete sequence, which anneals to its homologous gene strand and continues to elongate. We consider these “interruptions of extension” as random events which are unlikely to be repeated. We further tested the accuracy of our method by amplifying and examining fragments (< 600 bp) of the GAPDH genes, using our designed internal primers (F1_5′-TATTTTCAATCATTTGTTTC-3′, R1_5′-AATCATTGCAT CCGAGAACAA-3′; F2_5′-AACAGTTTAGTTGCCAATTCG-3′, R2_5′-CTCAATAC TGAAAATTTTG CTAG-3′). We applied this test to all of the Meconopsis species in this study and all of results confirmed that our method performed accurately.

After excluding PCR recombinants, we used the program Recombination Detection Program (Version RDP2) (Martin et al., 2005; available from  http://darwin.uvigo.es/rdp/ rdp.html) to exam natural recombination. We used the analysis algorithms RDP, GENECONV, Bootscan/Recscan, MaxChi, Chimaera, SiScan, and 3seq, which are implemented in RDP2. Detected natural recombinants were eliminated from further analysis.

Alignment and phylogenetic analysis – Alignments were performed using Geneious (Biomatters, New Zealand) Alignment 5.5 with the default settings, and then refined manually. The GAPDH sequences that lost multiple introns were eliminated from the final alignment. Phylogenetic analysis was carried out by RAxML 7.2.8 (Stamatakis, 2006; Stamatakis et al., 2008) with 1000 bootstrap replications.

Rooting of the tree – We obtained only one copy of GAPDH gene from the outgroup species Papaver alpinum which belongs to Papaver section Meconella, the sister group to Meconopsis. However, this sequence was difficult to align with Meconopsis sequences due to a great deal of sequence divergence and phylogenetic analyses using tentative alignments including this outgroup sequence could not resolve the relationship between the major clades with certainty (bootstrap value < 60 for the node directly shared by the outgroup and some Meconopsis species in unrooted trees). Therefore, we chose to use mid-point rooting.

Reconstructing polyploid evolution and ancestral chromosome number

The reconstructed cpDNA Bayesian tree of Meconopsis, Roemeria, and Papaver, rooted by Cathcartia, is shown in Fig. 2A, which is consistent with the Papaver phylogeny published by Carolan et al. (2006). The partial tree that contains only Meconopsis species is expanded in Fig. 2B.

Fig. 2A.

The cpDNA Bayesian trees of Meconopsis, Roemeriana, and Papaver rooted using Cathcartia with chromosome numbers of extant species at the branch tips. Ancestral chromosome numbers reconstructed using chromEvol are shown with the most probable chromosome number (2n with p value > 0.50) labeled above the tree branches and the posterior probabilities below the branches.

Fig. 2B. Reconstructed ancestral chromosome numbers for Meconopsis are shown above the branches and their posterior probabilities below the branches. Known chromosome numbers of extant species are shown at branch tips. Each section is labeled with a different color, corresponding to Fig. 3.

Fig. 3.

GAPDH phylogenetic network reconstruction in Meconopsis. Bootstrap value labeled above the branch when greater than 50. The three different patterns on the branches (left side) that lead to each of the clade (Clades 1-3) are used to symbolize the sequence origin for each clade and also corresponding to those in Fig. 4. Four colors indicate four sections: each of the colored lines (on the right) connects different copies of the same accession, revealing hybridization pattern; colored dots indicate the accessions with only one obtained GAPDH sequence.

Fig. 4.

Summary of ancient reticulate hypotheses in Meconopsis. (Note: this graph was transformed from Fig. 3. For example, sect. Aculeatae sequences have multiple origins from both Clades 1 & 3, represented by solid grey-patterned branch and dotted-line branch, respectively, in both Figs. 3 & 4).

Of the eight models implemented in chromEvol, the four that allowed DEMI (i.e., allowing chromosome number transition from 2n to 3n) all resulted in significantly higher likelihood scores than the other four models without DEMI setting (Table 1). The four DEMI models all generated the same ancestral chromosome numbers, which are shown in Fig. 2A and Fig. 2B. In Fig. 2A, the ancestral chromosome number of the genus Meconopsis is 2n = 14 (posterior probability 0.99). The ancestral chromosome numbers for each section, as shown in Fig. 2B, are 2n = 21 for M. sect. Grandes, 2n = 22 for M. sect. Primulinae, 2n = 28 for M. sect. Meconopsis, 2n = 14 for M. sect. Aculeatae. The chromosome number of the most recent common ancestor shared by M. sect. Grandes and M. sect. Primulinae was estimated to be 2n = 21.

Reconstructing GAPDH phylogenetic network

We successfully obtained partial GAPDH sequences from 21 Meconopsis accessions and one outgroup accession of Papaver alpinum. Because M. sect. Primulinae species are distributed narrowly and endemically, there were few good quality samples for successful PCR amplification. Thus, M. sect. Primulinae is less well represented in the GAPDH network than other sections. On average, 44% of the raw sequences from one PCR reaction were identified as PCR-mediated recombinants. One natural recombination in the sample X022 was detected by program RDP2. After the removal of the recombined sequences, we obtained an average of 2.1 copies of GAPDH gene per accession (1.3 in M. sect. Meconopsis; 2.3 in M. sect. Grandes; 2.5 in M. sect. Aculeatae; 2.0 in M. sect. Primulinae) to reconstruct the network. The average sequence length was 1403 bp with 885 variable sites and an intron/exon ratio of 3.4/1. The ML tree with best likelihood score is shown in Fig. 3 with the bootstrap value labeled above the branch when greater than 50. Three well supported major clades (labeled Clade 1, 2, and 3) are indicated for the ease of discussion (Fig. 3) and each is highlighted by a uniquely patterned branch (left in Fig. 3). Clade 1 comprises only species in M. sect. Aculeatae; Clade 2 contains species of M. sect. Meconopsis, M. sect. Grandes, and M. sect Primulinae; and Clade 3 includes sequences from M. sect. Aculeatae, M. sect. Grandes and M. sect. Primulinae. The network shows that GAPDH sequences in each M. sect. Aculeatae, M. sect. Grandes, and M. sect. Primulinae have multiple origins. We consequently transformed Fig. 3 to the simplified reticulate structure shown in Fig. 4 in order to illustrate our proposed scenario of ancient reticulate evolution in Meconopsis. The patterned branches in Fig. 4 correspond to those in Fig. 3.

Hybridization and Allotripolyploid evolution in Meconopsis

According to the ancestral chromosome reconstruction (Fig. 2B), there was an ancestor, with 2n = 21 (marked by the ★ in Fig. 2B) that gave rise to two extant ploidy groups in Meconopsis, M. sect. Grandes and M. sect. Primulinae. In the phylogenetic network using the low-copy nuclear marker GAPDH (Fig. 3 in which each Meconopsis section is indicated by a different color) the GAPDH sequences are clustered into three groups labeled Clades 1, 2, and 3. It is obvious that sequences of M. sect. Grandes and M. sect. Primulinae have multiple origins, located in both Clade 2 and Clade 3 on the GAPDH network (Fig. 3), which strongly disagrees with the cpDNA tree topology where each section is monophyletic (Fig. 2B). One explanation for this pattern of disagreement is that M. sect. Grandes and M. sect. Primulinae had hybrid origin(s). Under this hypothesis, Clade 2, similar to the cpDNA tree (Fig. 2B), could contain the maternal lineage and Clade 3 could represent the paternal lineage for both M. sect. Primulinae and M. sect. Grandes (Fig. 3). In addition, M. sect. Primulinae is morphologically most similar to M. sect. Aculeatae according to traditional classifications of Meconopsis (Prain, 1915; Taylor, 1934) but it is most distant from M. sect. Aculeatae on the cpDNA tree. A hybrid origin could also explain this morphological divergence pattern as hybrid offspring can sometimes predominantly display the traits of one of its parents.

Given our finding that Meconopsis sect. Primulinae and M. sect. Grandes shared a common ancestor of 2n = 21 (★ in Fig. 2B), the formation of this triploid ancestor was probably the result of a hybridization event, which fits the phylogenetic pattern displayed in Fig. 3. Alternatively, it is possible to hypothesize that two or more independent ancient hybridizations in M. sect. Primulinae and M. sect. Grandes respectively could also account for the network structure shown in Fig. 3. However, because, as shown in Fig. 2B, the ancestral chromosome numbers in M. sect. Meconopsis and M. sect. Aculeatae are (2n) 14, 28 or 56, we think hybridization between these cytological types with any of the deviants from a triploid (2n = 21) is unlikely.

It is also possible that the structure of GAPDH network could also be explained by gene duplication and gene loss. However, we considered this a less likely scenario because it would have to require multiple gene-loss events in the early history of the genus to result in the pattern of the observed GAPDH network. Also, the hybridization hypothesis fits the incongruence between nrITS tree and cpDNA tree observed from our phylogenetic study of Meconopsis (Xiao, 2013). However, this alternative hypothesis should be tested further in the future.

We therefore reiterate that the most likely scenario is that Meconopsis sect. Grandes and M. sect. Primulinae shared a triploid ancestor of 2n = 21 with a 2n = 14 maternal parent (evident when tracing back along the cpDNA phylogeny in Fig. 2B). This triploid ancestor (★ in Fig. 2B) appeared to be transient and later established stable lineages (with even-numbered chromosomes) through different putative pathways as follows:

Thus our results suggest that the emergence of two Meconopsis sections was attributable to a single ancient triploidization event. However, mechanisms of how this triploid hybrid ancestor successfully bypassed the low fertility and reproductive instability associated with aneuploidy and hybrid sterility (Comai, 2005) are unclear. It has been suggested that some reproductive strategies, e.g., apomixis, or being perennial, can prolong the life cycle and could contribute to higher chances of establishing reproductively stable lineages (Ramsey & Schemske, 1998). Presumably because a polycarpic life cycle (longevity) can increase chances of the occurrence of a mutation that might lead to eventual reproductive success, the polycarpic habit is frequently found associated with allopolyploidy while diploid populations of the same species are predominantly monocarpic (Treier et al., 2009). In Meconopsis, the majority of species are strictly monocarpic; polycarpic plants are only found in Meconopsis sect. Primulinae (i.e., M. bella) and M. sect. Grandes (i.e., M. betonicifolia, M. grandis). The long-lived polycarpic habit might be derived from, or related to, an allotriploid origin.

According to McNaughton (2014), fertility was gained over time in Meconopsis ‘Lingholm’, a cultivar produced by human hybridization between two species with different ploidy levels – apparently followed the sequence of steps given below:

McNaughton's report (2014) documented the formation of a new hexaploid (M. ‘Lingholm’) via a sterile triploid (M. × sheldonii), whose descendant, a few decades later, started to produce viable seeds with double the chromosome numbers of its progenitor. McNaughton (2014) commented that Meconopsis triploid cultivars were not uncommon in gardens but that the conversion of a triploid into a fertile hexaploid, as from M. × sheldonii to M. ‘Lingholm’, is a unique event, thanks to which, gardeners can now grow their own “blue heaven” (M. ‘Lingholm’). Our results that recovered an ancient triploid viewed in light of the evidence from M. ‘Lingholm’, suggest that polyploidy was a significant evolutionary force in Meconopsis and that polyploidization through triploidy may have occurred more than once through Meconopsis history. However, the mechanisms (e.g., somatic doubling, fusion of unreduced gametes) of a triploid's transition to a stable hexaploid or dodecaploid (as in M. sect. Grandes) a still largely remain hypothetical (Kadereit, 1987; Bretagnolle & Thompson, 1995; Ramsey & Schemske, 1998; Rieseberg & Willis, 2007; McNaughton, 2014).

In another relevant study, Kadereit (1986, 1987) presented the plausibility of a triploid origin in Papaver somniferum (2n = 20, 22, 44). He proposed a triploid origin of P. somniferum by providing several different lines of evidence supporting that P. somniferum was derived from a triploid hybrid (2n = 21). A later molecular study (Nessler, 1994) analyzing the MLP (major latex proteins) gene family supported an hypothesis of a triploid hybrid origin of the opium poppy. According to Kadereit (1991), the triploid overcame low fertility and stabilized as even number ploidy through regular bivalent formation. He (1987) also suggested that 2n = 22 is an indication of how a triploid (2n = 21) establish even-numbered ploids in Papaver and its closely related genera based on the chromosome number records of a few unrelated species sharing the same aneuploid number n = 11: Meconopsis bella (2n = 22), Papaver somniferum ssp. somniferum (2n = 20, 2n = 22), Papaver somniferum ssp. setigrum (2n = 44), Papaver aculeatum (2n = 22), Roemeria hybrida (2n = 22), Papaver cambricum (2n = 14, 22, 28). The high likelihood scores among the different models calculated by chromEvol (Table 1) with the DEMI parameters (allowing the genome transition from 2n to 3n) support Kadereit's (1986, 1987) conclusion, suggesting that these processes are probably not random and that Meconopsis-Papaver species are prone to overcome a triploid block and establish stable lineages by dysploidy formation (2n = 21→2n = 20/22).

Autoploidy within Meconopsis

In addition to Meconopsis sect. Primulinae and M. sect. Grandes, sequences of M. sect. Aculeatae also fell into two different clades (Clade 1 and Clade 3, Fig. 3), suggesting that M. sect. Aculeatae might also have a hybrid history. The reconstructed polyploid evolutionary scenario suggests that Meconopsis sect. Aculeatae and M. sect. Meconopsis acquired chromosome levels of 2n = 56 independently, which is strongly supported by the aberrant chromosome number 2n = 28 in M. sect. Meconopsis and 2n = 14 in M. sect. Aculeatae (Fig. 2B). This conclusion contradicts a previous assumption, postulated without any available phylogenetic reference at that time, by Ratter (1968) that species in M. sect. Aculeatae and M. sect. Meconopsis were derived from a common ancestor with 2n = 56. However, sequences of M. sect. Meconopsis are only in Clade 2 with no strong divergence as shown in other Meconopsis sections, thus, it is possible that the high ploidy level in M. sect. Meconopsis resulted via autoploidy. For easy visualization, we simplified GAPDH network and displayed it in a reticulate network form (Fig. 4), which illustrates and summarizes our hypotheses of reticulate evolution of Meconopsis.

In conclusion, we postulate that there were three major pathways early in the evolution of Meconopsis: (1) formation of a triploid hybrid that managed to overcome sterility at the base of the clade that gave rise to M. sect. Primulae and M. sect. Grandes; (2) a hybrid origin of M. sect. Aculeatae with successive polyploidizations; (3) a autoploidization origin of M. sect. Meconopsis. Future research will be needed to test these hypotheses further, perhaps using next generation sequencing technique to generate more nuclear sequence or transcriptome data. Nonetheless, this is the first hypothesis of ancient hybridization and polyploidy underlying the evolution of Meconopsis.