Species circumscriptions and relationships in the Racomitrium canescens complex

All eight morphologically defined species of the R. canescens complex plus R. varium are easily distinguishable from each other based on the combined rps4-trnT-trnL and ITS sequence data (Fig. 2). Much confusion about species delimitations in the R. canescens complex arose from early taxonomic attempts to subdivide R. canescens s.l. into various varieties or forms (overview in [17]), which were often based on environmental modifications of morphological characters only. Frisvoll [17] emphasized the frequent presence of mixed populations (‘mixed stands’) of morphologically distinguishable plants, which indicated that a number of genetically distinct taxa do exist in the R. canescens complex. In fact, the species formerly treated as varieties of R. canescens, namely R. barbuloides (R. canescens var. epilosum H. Müll. ex Milde, fide [20]), R. ericoides, and R. muticum [17], [18], and the species newly described by Frisvoll [17], viz. R. elongatum and R. pygmaeum, are molecularly clearly separated from R. canescens s.str.

Despite low phylogenetic distances in the R. canescens complex, genetic differentiation between species is generally larger than within species, as inferred from the generally smaller intraspecific than interspecific distances (Fig. 1, branch lengths in Fig. 2) and the significant pairwise Fst comparisons for five species of the complex (Table 2). It has to be noted, however, that the Fst estimates are based on a small sampling and only a small part of the genomes with few haplotypes. Furthermore, no incongruence between the plastid and ITS sequences was found, rejecting the occurrence of hybridization in the present dataset. The overlap in the total sequence distance ranges (Table 1) is due to an exception of the general observation of smaller intraspecific than interspecific distances, as the molecular divergence in ITS between R. muticum and R. pygmaeum is smaller than the intraspecific divergence within R. japonicum. It could therefore be argued that the former two should be treated as one species, whereas R. japonicum should be split into two. However, R. muticum and R. pygmaeum are easily distinguishable by morphological characters [15]–[18]. Racomitrium japonicum, in contrast, shows only little morphological variation, and if split, distinguishing the two segregate taxa might be only possible by molecular characters and possibly a geographic separation (China versus Japan, cf. Fig. 2). This needs to be tested by analyzing a larger number of specimens from the entire East Asian distribution area of R. japonicum.

Based on the present inferences we argue that the molecular data support Frisvoll's [17] thorough revision and that his morphological species circumscriptions should be maintained, as they correspond to well-supported clades in the molecular phylogenetic reconstructions. Molecular species delimitation in the R. canescens complex is thus straightforward, in contrast to several other genera of liverworts and mosses analyzed recently, where the presence of para- or polyphyletic species, cryptic speciation, incongruence between molecular markers, or incongruence between molecular and morphological characters complicated species delimitation (e.g. [2], [4], [21]–[24]).

In line with distinguishing the taxa of the R. canescens complex at the species level, intraspecific molecular divergence could be treated taxonomically at the subspecies level, especially when groups of specimens form at least moderately supported subclades, such as within R. canescens, R. ericoides, and R. japonicum (Fig. 2). In fact, one of the subclades within R. canescens corresponds to R. canescens subsp. latifolium (Fig. 2), which is morphologically and geographically separated from R. canescens subsp. canescens, although both morphological characters and the distribution range overlap between the two subspecies [17]. The taxonomic status of other intraspecific clades that are molecularly distinguishable within the widespread Racomitrium species remains to be tested. As in R. japonicum, subclades in R. canescens seem to be geographically separated (circum-North Pacific, western North America, Europe), which would provide support for treating them as separate subspecies in addition to subsp. latifolium, but more specimens, especially from outside Europe, need to be analyzed.

At the supraspecific level, the present molecular data do not support the division of the R. canescens complex into two groups of species, which were formerly treated as Racomitrium subsections Canescens and Ericoides [17] or Niphotrichum sections Niphotrichum and Elongata [18], respectively. According to these classifications, R. canescens s.str. and R. panschii should be closely related (both placed in subsect. Canescens) and separated from the remaining species of the complex (classified in subsect. Ericoides). However, the present molecular data indicate a closer relationship of R. panschii with R. barbuloides, R. japonicum, and R. varium (Fig. 2). Racomitrium japonicum, which was considered by Frisvoll [17] as distantly related to the remaining taxa of the R. canescens complex, is found here closely related to R. barbuloides (Fig. 2), with which it frequently grows together in mixed stands [17]. Both R. barbuloides and R. japonicum are predominantly East Asian species, the latter reaching southwards to Vietnam and Australia [25], [26]. Similarly, a close relationship is indicated between the circum-North Pacific R. muticum, which is most frequent in western North America [18], and the western North American endemic R. pygmaeum. Further inferences on species relationships as well as analyses of phylogeographic patterns in the R. canescens complex need further study based on a denser sampling especially in the North American-East Asian region and sequencing of additional markers to increase support for the backbone of the phylogenetic reconstruction (Fig. 2).

The clear molecular distinction between the species of the R. canescens complex indicates that the morphological characters used to identify them (leaf shape, hairpoint structure and stance, nerve length, alar cells and basal marginal leaf cells), albeit being in part rather small and difficult to apply, are reliable for species identification. On the other hand, misidentification of collections seems to be a severe problem in the R. canescens complex. If the molecular phylogenetic reconstructions accurately reflect species delimitations, 14 out of the 70 newly sequenced specimens (20%) were misidentified based on morphological characters (Figs. 2, 3). This could partly be due to mixed collections or field determinations without later checking the material microscopically (e.g. collections provisionally named R. canescens [sensu lato] to indicate that they belong to the complex), but might also be an inherent problem of the diagnostic characters. The majority of the 14 misidentified specimens belonged to species of subsection Ericoides sensu Frisvoll [17], not to R. canescens s.str. or subsp. latifolium (Fig. 3). Morphological re-investigation revealed that some of the collections named R. canescens, but actually belonging to R. ericoides or R. elongatum, were intermediate in terms of the diagnostic characters separating these three species. They showed rather broad and obtusely keeled leaves, typical for R. canescens, but a single, non-forked nerve reaching at least ¾ up the leaf, typical for the two latter species, members of subsect. Ericoides. The suitability of leaf shape as a diagnostic character hence is questionable, but this needs to be confirmed by analysis of a larger number of collections labelled R. canescens.

Download:

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

Figure 3. Revised species identifications in the Racomitrium canescens species complex based on molecular data.

Changes in species identification of 14 specimens analyzed in the present study are indicated by arrows. Arrow thickness is equivalent to the number of specimens transferred from one species to another (one, two, or four specimens, respectively). Grey or black arrows indicate changes within or between subsections Canescens and Ericoides [15], respectively.

https://doi.org/10.1371/journal.pone.0053134.g003

Implications for DNA barcoding of species complexes in mosses

Species identification by DNA barcoding seems to be more difficult in bryophytes than in other land plant lineages or in animals. Instead of using a single, short piece of DNA to discriminate species across a wide range of lineages, barcoding in land plants is supposed to be based on one or two core markers, plus additional information from other DNA regions where necessary (e.g. [11], [27], [28]). Different marker combinations, mostly from the plastid genome, have been proposed (see [11] for review), such as rbcL and matK as core markers for land plants [5], or the psbA-trnH spacer together with the nrITS region for angiosperms [29]. In bryophytes, especially mosses, however, the respective plastid markers either tend to be short (psbA-trnH spacer, trnL-F spacer, [30], [31]), their discrimination capacity at the species level is still debated (rbcL, [8], [10]), or efforts are still needed concerning primer design and amplification strategy (trnK/matK, e.g. [7]). A stand-alone barcoding marker is unlikely to be found among the standard plastid markers used for phylogeny reconstruction [11]. Liu et al. [8] identified five plastid markers with amplification success >90% and taxon assignment success >80% (in different combinations up to 92%), viz. rbcL, rpoC1, rps4, psbA- trnH, and trnL-F. The suitability of these markers for discriminating closely related moss species in species complexes, however, remained difficult to assess, as only few species from selected genera were compared. The subsequent study focusing on Grimmiaceae [9] already showed that (partial) rbcL sequences did not perform well, and the discrimination capacity of the best marker (psbA- trnH) did not exceed 65% in the whole family. In Racomitrium, all 11 species included in Liu et al. [9] could be distinguished by rps4, and with >80% success by rbcL and psbA-trnH, but again the question arises whether these markers would provide sufficient information when further closely related species are compared.

In the present study of the Racomitrium canescens complex, almost no differences were observed between the plastid rps4-trnT-trnL and nuclear ribosomal ITS regions concerning amplification and sequencing success. Both regions could be sequenced completely for at least 90% of the analyzed specimens. The ITS region showed a clear gap between intra- and interspecific sequence divergence for the species represented by more than one accession (Fig. 1) and allowed to resolve clades of all species of the complex with significant statistical support (Table 3). The same was true for ITS1 alone (the clade of R. varium, R. barbuloides and R. japonicum was not resolved, but R. varium still formed a clade separate from the other species). Although the different parts of the rps4-trnT-trnL region and the ITS2 performed well, too (Fig. 1), they failed to discriminate R. pygmaeum and R. muticum (Table 3). Including indels by simple indel coding did not significantly improve discrimination capacity of any marker and in some cases even resulted in lower clade support (Table 3). Here a different indel coding strategy might be desired.

Advantages and disadvantages of using the nrITS region as DNA barcode have been discussed, e.g., by [29], [32]. Apart from inherent problems such as the possible presence of paralogous ITS copies or incongruence between ITS1 and ITS2 (e.g. [33]), amplification difficulties of the whole ITS1-5.8S-ITS2 region need to be taken into account. The latter, however, can often be solved by amplifying ITS1 and ITS2 separately, which has also been done in phylogenetic analyses of bryophytes (e.g. [34]). Until now, mostly ITS2 has been considered as plant DNA barcode marker [32], [35]. Bell et al. [7] considered the whole ITS region for the liverwort genus Herbertus. However, amplification and sequencing success was lower than in the employed plastid markers. With respect to the results of the present study, we argue that ITS1, which is generally more variable than ITS2 [36], should also be considered as potential barcode marker in complexes of closely related species of mosses. The preliminary observation in the Racomitrium canescens complex that ITS1 seems to outperform ITS2 should be tested in further species complexes to decide which part of the nrITS region works best as potential core barcoding marker for species identification in mosses. Furthermore, DNA barcoding in difficult species complexes should be complemented by redefining morphological characters, to develop a set of easy-to-use molecular and non-molecular identification tools for improving biodiversity assessments and ecological research, including mosses.

Ethics statement

All necessary permits were obtained for field studies to collect material used for molecular analysis in Svalbard (Governor of Svalbard, references 2008/00688-2, 2009/00412), Greenland (BioBasis project, Dr. N.M. Schmidt, Aarhus University), Madeira (Dr. S. Fontinha, National Park of Madeira Services/Madeira University), Azores (Secretaria Regional da Agricultura e Florestas Dos Açores), UK (H. Cole, Manager/Senior Ranger Naturalist, Ben Lawers National Nature Reserve), and The Netherlands (Dr. H. van der Hagen, Dunea N.V.). Further material was collected in areas without specific permissions required or concerned herbarium collections from G, KRAM, L, and S, which were used according to regulations of the respective herbaria. Areas without permission needed were neither privately owned nor protected. The species sampled are neither protected nor listed by CITES (Convention on the International Trade in Endangered Species).

Taxon sampling, DNA extraction, PCR and sequencing

The present taxon sampling and outgroup selection was based on recent molecular phylogenetic reconstructions of Racomitrium s.l., which showed that R. fasciculare and R. laevigatum (both treated as Codriophorus by [13]), formed a sister clade to the R. canescens complex [15], [19] (treated as Niphotrichum by [13]), the latter including R. varium [15] (also treated as a Codriophorus in [13]). Sequences from the plastid rps4-trnT-trnL and nuclear ribosomal ITS regions were newly generated for 70 specimens of the Racomitrium canescens complex. Voucher information and Genbank accession numbers are listed in Appendix S1. Six further collections could not or only partially be sequenced or the sequences were of insufficient quality. In addition, sequences of five Racomitrium specimens were taken from earlier studies, namely R. elongatum (Spain) from [14] and R. ericoides (both samples from Poland) as well as R. fasciculare and R. laevigatum as outgroup representatives from [19].

Distal parts of single shoots were thoroughly cleaned with distilled water. Total genomic DNA was extracted using the DNeasy® Plant Kit (Qiagen) or the NucleoSpin® Plant II Kit (Macherey-Nagel). The employed molecular markers were amplified by PCR using protocols and primers as described in [14] for rps4-trnT-trnL (primers rps4-166F and P6/7) and [37] for nrITS (primers 18F and 25R). In a few cases of difficulties with obtaining PCR products, the rps4-trnL part was split into two halves, which were amplified and sequenced separately with primers rps4-166F and A-Rbryo or A-Fbryo and P6/7, respectively [14]. PCR products were purified using the Wizard® DNA Clean-up kit (Promega) or by Macrogen Inc. (www.macrogen.com), where the automated sequencing was performed as well. Sequencing primers were those used for PCR.

Alignment, sequence analysis and phylogenetic reconstructions

DNA sequences were manually aligned in PhyDE® v0.995 [38]. Phylogenetic reconstructions of taxon circumscriptions and relationships were performed based on the maximum parsimony (MP) principle and using two model-based approaches, maximum likelihood (ML) and Bayesian inference (BI). Separate MP analyses of the rps4-trnT-trnL versus ITS sequences were performed and the resulting tree topologies were checked for possible incongruence between the plastid and nuclear markers by visual inspection and by applying a partition homogeneity test (ILD test) [39], [40] as implemented in PAUP* 4.0b10 [41] (100 replicates). Calculations of pairwise Fst estimates and p-values based on plastid and ITS haplotypes were performed using Arlequin v3.5.1.3 [42], with haplotypes (including indel characters) delineated in TCS v1.2.1 [43].

Phylogenetic analyses of the combined plastid and nuclear markers were run using MP, ML, and BI. To evaluate the employed markers for DNA barcoding, further MP analyses were performed of four partitions for which primers pairs are available, viz. rps4-trnT, trnT-trnL, ITS1, and ITS2. In addition, pairwise nucleotide distances between all sequences were calculated according to the K2P model (cf. [8], [9]) for the combined dataset and all partitions, and compared between and within the species of the R. canescens complex.

Calculation of pairwise distances as well as MP and ML analyses were performed in PAUP. Heuristic searches under parsimony were implemented using random sequence addition with 1000 replicates and tree bisection-reconnection (TBR) branch swapping. All MP analyses were run with gaps (indels) either treated as missing data or coded as informative by a simple indel coding (SIC) strategy [44] as implemented in SeqState [45]. Heuristic bootstrap searches under parsimony were performed with 1000 replicates and 10 random addition cycles per bootstrap replicate with the same options in effect. To search the tree space for islands of more parsimonious trees, parsimony ratchet analyses were performed with PRAP2 [46] in combination with PAUP, employing the default options (200 iterations, 25% of randomly chosen positions up-weighted to 2) and superimposed 10 random addition cycles.

For the model-based approaches, model testing was performed in Modeltest 3.7 [47] employing MrMTgui [48]. GTR+Γ+I was indicated as best-fit model of the combined dataset according to the Akaike information criterion (AIC). Consequently, the settings basefreq = (0.3011 0.2002 0.2254), nst = 6, Rmat = (1.5235 5.1208 0.6544 2.3313 7.1000), rates = gamma, shape = 0.7849, and pinvar = 0.6263 were used for ML, and nst = 6 and rates = invgamma for BI. Bayesian posterior probabilities were calculated based on the Metropolis-coupled Markov chain Monte Carlo (MCMCMC) method, using MrBayes v3.1 [49]. In a second set of Bayesian analyses the indels coded by SIC were included, with sequence and indel data treated as separate and unlinked partitions, employing the restriction site model (‘F81’) for the indel matrix. The a priori probabilities supplied were those specified in the default settings of the program. Four runs with four chains (10⁶ generations each) were run simultaneously, with the temperature of the single heated chain set to 0.2. Chains were sampled every 1000 generations and the respective trees written to a tree file. Fifty percent majority rule consensus trees and posterior probabilities of clades were calculated by combining the four runs and using the trees sampled after the chains converged. Trace plots generated in Tracer v1.5 [50] were used to check for convergence of the runs (plateaus of all runs at comparable likelihoods) and to infer the ‘burnin’, which ranged approximately between the first 100000 and 120000 generations (first 100–120 sampled trees). Consequently, the first 150 trees (15%) were deleted to be sure that only trees of the stationary phase were included.