Ethics statement

The plant sampling of this study includes 21 accessions derived from Yu et al. [20] and 41 accessions borrowed from herbariums of East China Normal University (HSNU) and Royal Botanical Garden Edinburgh (E) (Table S1). No specific permission for field work was needed for this study and the sampling was carried out without destroying local populations.

Molecular phylogeny

We sampled a total of 59 accessions representing the ingroup (Table S1), including Cololejeunea lanciloba (8 accessions), C. planissima (23 accessions.), C. latilobula (14 accessions), and C. yakusimensis (8 accessions). In addition, the ingroup also comprised two accessions each of two other members of C. lanciloba complex (C. cocoscola and C. thailandensis) and one accession each of two relatively closely related taxa (C. japonica (Schiffn.) Mizut. and C. stylosa (Steph.)A. Evans).The outgroup comprising three accessions of Cololejeunea calcarea (Libert.) Schiffn. was chosen based on the phylogeny of Cololejeunea [20]. The key morphological data of four MTSs of interest was assembled based on the description from literatures [15,35,36] and observations on available herbarium specimens (see Table 1). These studies should aim to include DNA sequence data of all type specimens as far as the preservation of the specimens enable DNA amplification. Each specimen studied was carefully identified using morphological evidence (see Table 1) before and after phylogenetic analyses. The chloroplast trnL–F and nuclear regions nrITS were obtained for all 62 accessions as described [20] or by downloading from GenBank (voucher information see Table S1). Sequences were aligned using PhyDE v0.9971 (http://www.phyde.de/). Ambiguous positions, positions for which at least two equally probable alignments were obtained, were identified visually and removed from alignment for subsequent analyses. Exclusion of ambiguous regions, the dataset was reduced from 1620bp to 1464bp (ca. 90% of the original matrix). The data matrices and a single tree of Maximum likelihood analyses (Figure 1) are deposited on TreeBase (URL: http://purl.org/phylo/treebase/phylows/study/TB2:S14897). All phylogenetic and network analyses were carried out for both regions independently and the combined dataset.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 1. The most likely phylogram obtained in a maximum likelihood analysis of the combined dataset.

Bootstrap values ≥ 95% (** = 100%, * = 95-99%) were plotted above branches and posterior support values ≥ 0.95 (**=1.00, *=0.95-0.99) below branches. Shown branch lengths correspond to the estimated substitution events (see bar). Vertical bars on the right side of the figure mark specimens that were recovered as part of a single network in statistical parsimony analyses of the combined dataset. The numbers correspond to the number of haplotypes per network. Letters on the right side correspond to molecular operational taxonomical units (MOTUs) discussed in the text. The amplification numbers and region are given for each specimen.

https://doi.org/10.1371/journal.pone.0084124.g001

Maximum likelihood analyses (ML) for separate regions and a combined dataset were conducted with PhyML 3.0 [37] as implemented in the PhyML plugin of Geneious 5.5.6 (http://www.geneious.com/). The optimal models of evolution were selected in jModeltest 0.1.1 [38] using Akaike Information Criterion (AIC) and hierarchical likelihood ratio test (hLRT): TIM + I for trnL–F, TrN + I + G for nrITS, and TrNef + G for the combined dataset. Bootstrap values for ML analyses were obtained via 200 bootstrap replicates in PhyML using the above fit model and parameters as in the optimal tree search. Combinability of the plastid and nuclear regions were inferred by comparing the bootstrap consensus trees, considering the 70% bootstrap value criterion as indicator of topological heterogeneity [39]. Bayesian inference of phylogeny for the combined dataset was performed using MrBayes 3.2.1 [40], and a partitioning into chloroplast region (trnL–F) and nuclear region (nrITS). Bayesian search was carried out with the GTR model implemented, parameter values inferred simultaneously with tree searcher, runs starting with a random tree, unlinked rates. The model was selected using the Bayesian Information Criteria (BIC) implemented in jModeltest. Four Markov chain Monte Carlo runs consisted of five million generations with a sampling frequency of one every 1,000th generation. The convergence of runs and burn-in phase were estimated in Tracer v 1.4.4 (http;//beast.bio.ac.uk/Tracer). The first 500,000 samples were discarded as burn-in samples. Bayesian posterior probabilities (BP-PP) were calculated as majority consensus tree of the remaining 4,500,000 trees after discarding the trees sampled within the burn-in phase.

ALTER (http://sing.ei.uvigo.es/ALTER/) was used to transform the matrix into the format required for TCS and to identify the number of haplotypes. Plausible connections between haplotypes were reconstructed using statistical parsimony approach (SPNA) with a 95% connection limit as implemented in TCS v1.21 [41,42]. To minimize the number of ambiguously inferred haplotypes, indels and one sample with missing data, Cololejeunea planissima L13, were excluded from the alignment for SPNA.

Species distinctiveness measures

Rosenberg’s reciprocal monophyly P(AB) and Rodrigo’s (P(RD) measures were calculated using the species delimitation plugin [43] for Geneious (www.biomatters.com). Two approaches were applied to investigate the species distinctiveness. Firstly, we tested distinctiveness by P(AB) and P(RD) for seven molecular operational taxonomy units (MOTUs) out of the eleven MOTUs that were selected based on clade support and the independence of network on 95% connection limit in SPNA. If two sister clades both have branch bootstrap ≥ 95% and involved in individual network in SPNA, we suggested that these two lineages are isolated and distinct. Four MOTUs, each including a single haplotype prohibiting the calculation of the two criteria, were excluded. Secondly, the taxonomic distinctiveness was assessed with the iterative ‘tip-to-root’ approach [33] on phylogeny without any predefined MOTUs. In the “tip-to-root” process, we assigned a unique number for each clade (e.g. 1–1) comprising two or more individuals in the phylogeny of Cololejeunea lanciloba complex. Every group or clade was tested against its sister group to assess the significant of distinctiveness according to P(AB) and P(RD). This process starts at the tips of the tree and works along the branches.

Phylogenetic and statistics parsimony network analyses

The combined dataset included 1,464 (1,062 nrITS, 402 trnL–F) characters of which 51 (42 nrITS, 9 trnL–F) were autapomorphic, and 285 (239 nrITS, 46 trnL–F) parsimonious informative. No evidence for topological incongruence was recovered between the two regions (Figure S1). Identical topologies were found in the consensus tree of the Bayesian analyses of the combined dataset with a partition in two regions (mean likelihood of –InL = 5,453.208) and the most likely tree found in the maximum likelihood analyses (–InL = 5,606.879). All clades based on the combined dataset were also recovered in the phylogeny based on the nrITS. A majority of deep nodes lacked support in the phylogenetic hypothesis based solely on the trnL–F region (Figure S1).

The four MTSs investigated were not resolved as monophyletic groups (Figure 1). Specimens identified as Cololejeunea planissima were nested in four clades: A, D, J and K, whereas those identified as C. latilobula were nested in clade C, H and J. The polyphyly of C. lanciloba was attributed to two accessions, one from Australia (Clade I) and another one from Indonesia (Clade F), distinct from clade E comprising of Chinese C. lanciloba specimens (Figure 1). Accessions of C. yakusimenis were nested in clade J, which also included samples identified as C. latilobula and C. planissima.

A total of 20, 36 and 38 haplotypes were found within the ingroup based on trnL–F, nrITS and the combined data set respectively. SPNA analyses recovered 7 independent networks and 4 isolated single haplotypes based on the combined dataset (Figure S2). In the phylogenetic analyses, the clades corresponding to these independent networks obtained high bootstrap values (ML-BP ≥ 95%) and posterior probabilities (BP-PP ≥ 0.95) (Figure 1; Table 2).

Species distinctiveness measures

Seven predefined MOTUs (clade A, C–E, H, J and K) were tested using P(AB) and P(RD) species distinctiveness measures. Under strict criteria, the mean probabilities (P ID (strict)) ranged from 0.55 (0.40 to 0.70) for clade D to 0.94 (0.89 to 0.99) for clade A (Table S2). When using more liberal criteria, the mean probabilities for all MOTUs investigated were above 90% (Table S2). Clade D and K have P(RD) values < 0.05, whereas clade A and clade J have P(AB) values < 10^-5. The clades D, E and H did not show genetic distinctiveness at significant level of P (AD) < 0.05 or P(AB) < 10^-5 despite they were found with support values of ML–BP = 100% and BP–PP =1.0 (Figure 1; Table S2).

The ‘tip-to-root’ approach recovered six additional MOTUs nested within the predefined seven MOTUs (Figure S3; Table S3, Table 2). These six additional MOTUs were: one each in clade A and E, and two each in clade C and J (Table 2). The results of species delimitation based on nrITS were similar with those based on the combined dataset (Table 2). A total of five additional MOTUs were revealed based on ITS: one each in clade A, E and J, and two in clade C (Table 2). The trnL–F data supported total seven MOTUs: one each in clade A, C, H, J and K, and two in clade E excluding untested clades (Figure 1; Table 2).

Overall, the analyses recovered evidence for distinct 13 MOTUs using P(AB), P(RD), clade support and SPNA, after excluding untested MOTUs comprising a single haplotype: one each in clade D, H and K, two each in clade A and E, and three each in clade C and J (Table 2).

Cryptic species within Cololejeunea lanciloba complex

Our results revealed that the monophyly of each MTS of interest were not supported by genotypic evidence. This conclusion was supported from both the nuclear and the plastid genome. In turn, we identified up to thirteen genetic distinct MOTUs among four MTSs investigated using statistical criterias P(AB) and P(RD). Specimens of Cololejeunea planissima, C. latilobula and C. lanciloba were nested in more than one clade separately, suggesting the discordance of phenotype and genotype. This conflict between phenotype and genotype was evident in particular for Cololejeunea planissima (clade 1-9, 1-8, D, K) and C. latilobula (clade1-13, 1-10, 1-11, H). The recovered topology indicated the occurrence of cryptic species. All accessions of C. yakusimensis together with some accessions from C. latilobula and C. planissima were nested in one clade J (Figure 1). These results suggested the occurrence of morphological plasticity.

Incongruence between morphological and molecular evidence has been reported in various plant lineages [44,45], including mosses and liverworts [46-49]. Several mechanisms considered to create these conflicts have been widely recognized including the occurrence of cryptic speciation. The origin of some cryptic species cases coincide either with extrinsic factors, e.g., ecological selection, or with inherited constraints, such as development constraints and functional constraints [2], whereas some cryptic species are the result of hybridization [50], and/ or introgression [51] and/or chromosome doubling [52].

Natural selection was considered to foster cryptic species via direct and/or indirect selection on phenotype and/or reproductive traits [6,53,54]. In our case, epiphyllous liverworts likely experience strong selection on morphological characters that are critical for the survival of these plants in harsh condition-at the surface of living leaves. The putatively adaptive characters of epiphyllous liverworts include small size of the gametophyte body, inflated leaf lobules and/or presence of papillosae cells for the retention of water, appressed stems and secondary rhizoid discs supporting attachment to the smooth leaf surface, substantial asexual production and inbreeding supporting effective colonization of ephemeral substrates [55,56]. Observations on rheotypic relatives of epiphyllous liverworts, such as Colura irrorata (Spruce) Heinrichs, Y.Yu, Schäf.-Verw. & Pócs [57] and Myriocoleopsis [58,59] support this hypothesis because these taxa are not only different in their ecological preference but are also distinct from their epiphyllous relatives in having creeping stolons, robust stems, and long androecial spikes. These characters are interpreted as adaptive characters for successful survival under running water. The observation of morphological distinctiveness of non-epiphyllous and epiphyllous species of Colura and Myriocoleopsis supports the argument that stabilizing selection acts on phenotypes of epiphyllous liverworts. Such case is congruent with the hypothesis of the breakdown of ecological constraints [60]. Thus, we assumed that the parallel evolution and/or phenotypic plasticity of theses adaptive traits may occur frequently in leafy liverworts occurring in extreme habitats in which only a limited number of strategies enable survival. In context of above indicated evidence, we hypothesized that cryptic species of the Cololejunea lanciloba complex are caused by strong stabilizing selection imposed by environmental conditions.

Alternatively the limited morphological variation of Cololejeunea species may coincide with restricted developmental options provided by their rather simple body-plan [61]. Developmental constraints have been considered in bryophytes in the context of a disconnection between wide distribution of MTSs and molecular variation among populations [62]. Although we have insufficient knowledge of alternative functional constraints on phenotypic evolution of liverworts, habitat-driven morphological constraints may contribute to the occurrence of cryptic species in the Cololejeunea lanciloba complex.

Polyploidization and/or hybridization are common causes of cryptic speciation in vascular plants [63,64] and mosses [65-68], but only a few cases of hybridization and/ or polyploidization have been reported in liverworts. These reports are distributed randomly through the phylogeny of liverworts and include some genera: Ricca [69], Marchantia [70], Plagiochasma [71], Porella [72] and Pellia [73], but did not include epiphyllous liverworts with the putative exception of Trocholejeunea [74]. The absence of evidence may be a result of the limited cytological studies or special life cycle period-haploid gametophyte. The latter hypothesis was consistent with current karyological observation on Lejeuneaceae including Cololejeunea [75-77]. Besides, our results did not provide evidence of hybridization and/or introgression because we did not observe any topological conflict between nuclear and plastid regions. However, we hesitate to make a conclusion of absence of hybridization and/or introgression based on utilization of only two loci. Future studies will hopefully explore this question using multiple loci because they can be effectively identified using next generation sequencing.

DNA taxonomy and species delimitation of cryptic species

The significance of DNA sequences for the recognition of cryptic species has been widely accepted, since crossing experiments, considered as a final test for reproductive isolation [78], are impractical for the majority of organisms. This applies especially for organisms adapting to extreme habitats. However, DNA taxonomic approaches are still under development and require further empirical investigation, especially in the context of land plants [31,32,79]. The required number of loci to obtain a robust species hypothesis has been recognized as one of the main problems of DNA taxonomy [80]. Single-loci-based studies may result in misleading taxonomic conclusions because these analyses are likely more sensitive to processes such as incomplete lineage sorting, genetic drift and lateral transfer of genes through hybridization and introgression. Nevertheless, reliable estimates of species boundaries based on single or few loci were derived in simulation studies [33]. In the present study we used two of the most frequently sequenced loci in land plants: the nuclear region nrITS and the plastid region trnL–F. These loci are known to accumulate both inter- and intra-specific sequence divergence for liverworts lineages and provide information on two out of the three genomes of the plant cell, and the sequence evolution of these regions is widely regarded as neutral [81].

This sensitivity of P(AB) and P(RD) in species delimitation was also supported by our results (see Table 2). To our knowledge, this is the first study employing these criteria to delimit species in liverworts, although these or related criteria have already been used in several studies of animals [82,83]. Species recognition has been discussed in previous studies of liverworts using DNA data, but none of them introduced statistic criteria, instead of general criteria such as monophyly, paraphyly and genetic distance. Furthermore, our results supported the feasibility of the ‘tip-to-root’ approach [33] for species recognition. We also want to point out the possible recovery of morphological distinctiveness by exhaustive morphological studies such as geometric morphometrics [84], which may enable us to find morphological disparity in cryptic species.