Introduction

An aquatic plant called Süßwassertang, which means “freshwater seaweed” in German, has been commercially available on the aquarium market worldwide for a number of years (Fig. 1a). Because of its liverwort-like appearance, it has long been considered to be a liverwort, such as Pellia or Monoselenium. Our observation of Süßwassertang gametangia, which are only rarely produced by the submerged thallus, suggested that this plant is not a liverwort but a fern gametophyte. Its archegonia are fern-like in having short necks, and the venter is immersed partly in the thallus. The antheridia resemble those of polypodialean ferns in that they consist of three cells: a cap cell, a ring cell, and a basal cell (Fig. 1b; Nayar and Kaur 1971). Although gametangia are present occasionally, sporophytes have never been observed in aquaria and even after planting onto soil. It is therefore difficult to identify the plant with certainty.

Fig. 1
figure 1

Süßwassertang, its microscopic features, and the habit of Lomariopsis spectabilis. a A portion of the gametophyte thallus showing extensive lateral branching. Bar: 1 cm. b Side view of an antheridium, showing a cap cell (cc), ring cell (rc), and basal cell (bc). Bar: 20 µm. c Scanning electron microscope image of developing lateral branches with rhizoids (arrowhead) and meristems (m) in the rounded apex. Bar: 0.2 mm. d Ribbon-like, branched gametophyte (g) of L. spectabilis bearing a young sporophyte (sp) in a field of Taiwan. It has similar morphology with the mysterious gametophyte. Arrowhead Branch points. Bar: 1 cm

Full size image

To unravel this mysterious identity, we employed a DNA-based identification approach consisting of sequence comparisons and phylogenetic analyses. We sequenced five chloroplast regions [rbcL, accD, rps4–trnS, trnL intron, and the trnL-F intergenic spacer (IGS)], of which rbcL has already been successfully used to identify an unknown fern gametophyte in a similar study (Schneider and Schuettpelz 2006). The widespread occurrence of rbcL in online databases for all plant lineages makes it well-suited for broad-scale screening. Likewise, the analyses by Quandt et al. (2004) indicated that the trnL-F region (intron and IGS) has the power to relate any sequence via a database comparison on generic level in most land plant lineages. The rapidly evolving gene rps4 (plus rps4–trnS IGS) was included because it represents one of the broadly sequenced regions (together with the trnL-F region and rbcL) in seedless plants (e.g., Quandt and Stech 2003; Schneider et al. 2004). In the approach chosen here, BLAST results pinpointed which clade Süßwassertang belongs to and therefore guided the taxonomic sampling for phylogenetic inferences. To increase the phylogenetic signal, we combined accD with rbcL in the analyses. Once narrowed to a certain lineage, a more precise marker was then used to resolve the position of Süßwassertang among more recently diverged lineages. In this case, the plastid trnL-F IGS was employed to infer inter-species affinities of this supposedly aquatic fern gametophyte.

Materials and methods

Taxonomic sampling for molecular phylogeny

Taxonomic sampling was guided by the MegaBLAST (Zhang et al. 2000) results obtained from the five regions sequenced (rbcL, accD, rps4–trnS, trnL intron, and trnL-F IGS). To obtain more robust confirmation of the relationship of Süßwassertang, we compiled two data sets for phylogenetic analyses. As indicated by the BLAST results, the first data set comprised a representative set of sequences from two coding plastid regions, rbcL and accD, of polypod ferns. In addition to the Süßwassertang sequences, 32 species representing 28 fern genera were included in the analyses. Athyrium niponicum (Mett.) Hance and Matteuccia struthiopteris (L.) Todaro were used as outgroups. The second matrix included all available trnL-F IGS accessions of Lomariopsis plus sequences obtained from two Süßwassertang accessions, L. spectabilis (Kunze) Mett., and Cyclopeltis crenata (Fée) C. Chr. Cyclopeltis crenata and Hypodematium crenatum Kuhn ex v. Deck. were used as outgroups for the second matrix.

DNA extraction, amplification, and sequencing

Total genomic DNA was extracted using either the Plant Genomic DNA Mini kit (Geneaid, Taipei, Taiwan) or the Plant Genomic DNA Purification kit (GeneMark, Taichung, Taiwan). In some cases a modified CTAB (cetyl trimethylammonium bromide) procedure (Wang et al. 2004) was applied. The PCR amplifications, which followed standard PCR protocols, were performed in 50-μl reaction volumes containing 1.5 U Taq DNA polymerase, 1.0 mM dNTPs-Mix, 10× buffer, 1.5 mM MgCl2, 10 pmol of each amplification primer, and 1.0 µl DNA. The PCR primers used for amplification and sequencing were: trnL-F with primers C (or E) and F (Taberlet et al. 1991; modifications according to Quandt and Stech 2004); rps4 (plus rps4–trnS IGS) with rps5′ (Nadot et al. 1994) and trnS (Souza-Chies et al. 1997); rbcL with NM34 (Cox et al. 2000) and M1390 (Lewis et al. 1997); accD with the newly designed primers “FW_accDF” (5′-ACG TCT GTA ACA AAT TGG TTT GAA G-3′) and “FW_accDR” (5′-AAA CTC AAC GTT CCT TCT TGC AT-3′). The PCR products were either directly purified using the GeneMark PCR Clean-Up kit (Taichung, Taiwan) or cleaned via gel extraction employing the Nucleospin PCR Purification kit (Macherey-Nagel, Düren, Germany). Sequencing was done with the amplification primers by Macrogen (Seoul, Korea). In order to corroborate the results, isolation, amplification, and sequencing of all regions were performed independently on two different samples in Taipei and Dresden. Newly obtained sequences and other accessions from GenBank used in the analyses are summarized in the Appendix.

Sequence alignment and phylogenetic analyses

DNA sequences were manually aligned using PhyDE0.995 (Müller et al. 2005). During manual alignment, gap placement was guided by the identification of putative microstructural changes following recently published concepts (Kelchner 2000; Quandt et al. 2003). Identified inversions were positionally separated in the alignments, but they were included as a reverse complement in the phylogenetic analyses, as discussed in Quandt et al. (2003). Phylogenetic reconstructions using parsimony were performed using winPAUP* 4.0b10 (Swofford 2002) in combination with PRAP (Müller 2004). The latter program generates command files for PAUP* that allow parsimony ratchet searches as designed by Nixon (1999). In our study, ten random addition cycles of 200 ratchet iterations each were used, with 25% of the positions being randomly double-weighted. The shortest trees collected from the different tree islands were finally used to compute a strict consensus tree. Heuristic bootstrap searches (BS; Felsenstein 1985) were performed with 1000 replicates, ten random addition cycles per bootstrap replicate, and otherwise the same options in effect as in the ratchet.

For a further measurement of support, posterior probabilities were calculated using MrBayes V3.1 (Ronquist and Huelsenbeck 2003), applying the GTR + Γ + I model. The a priori probabilities supplied were those specified in the default settings of the program. Posterior probability (PP) distributions of trees were created using the Metropolis-coupled Markov chain Monte Carlo (MCMCMC) method and followed the search strategies suggested by Huelsenbeck et al. (2001, 2002). Ten runs with four chains (106 generations each) were run simultaneously. Chains were sampled every ten generations, and the respective trees were written to a tree file. Calculation of the consensus tree and of the PP of clades was performed based upon the trees sampled after the chains converged (within the first 250,000 generations). Consensus topologies and support values from the different methodological approaches were compiled and drawn using TreeGraph (Müller and Müller 2004).

Results

BLAST results of the sequenced markers

The sequences (rbcL, accD, rps4-trnS, trnL intron, and trnL-F) obtained from two independent collections of Süßwassertang were identical. BLAST results indicated that Süßwassertang shares high sequence similarities to leptosporangiate ferns and is closest to the Lomariopsidaceae (except for trnL intron), especially in terms of the reported maximum identity (Table 1). Lomariopsis lineata (C. Presl) Holttum was found to be the best match for the trnL-F IGS, whereas for the three coding regions (rbcL, accD, and rps4), L. spectabilis Mett. or L. marginata (Schrad.) Kuhn. received the highest maximum identity scores. Although members of the Dryopteridaceae were among the best matches in a BLAST search using trnL, these results are biased since this region is currently represented by only few ferns in GenBank.

Table 1 Results of BLAST searches in GenBank, with only the first ten hits shown
Full size table

Phylogenetic analyses

Phylogenetic inferences of a representative set of polypod ferns for each of the single-gene data sets (rbcL and accD) were congruent. The phylogenetic relationships presented are thus based on the analyses using the combined data set. Maximum parsimony and Bayesian inference both clearly positioned the aquatic fern gametophyte within the fern genus Lomariopsis (Lomariopsidaceae), a placement that receives high branch support in the phylogenetic analyses (BSMP = 100, PPMB = 1.0; Fig. 2).

Fig. 2
figure 2

One of two most parsimonious trees [length 1739 steps, consistency index (CI) 0.449, retention index (RI) 0.602, rescaled consistency index (RC) 0.270] retained by the parsimony ratchet analysis performed based on the combined rbcL and accD sequence data. This tree was chosen as it perfectly reflects the Bayesian inferences. The values above the branches refer to posterior probabilities from Bayesian analysis, whereas those below the branches indicate bootstrap support values

Full size image

Phylogenetic analyses of Lomariopsis based on the trnL–F IGS (340 nt) indicated that Lomariopsis lineata (C. Presl) Holttum is the species closest to Süßwassertang (BSMP = 96, PPMB = 1.0; Fig. 3). The trnL-F sequences from L. lineata and the aquatic fern gametophyte show a 97.6% similarity and share an 8-nt indel (Fig. 3) that is absent in other Lomariopsis species. Despite the strong sequence similarity in the non-coding region of the chloroplast genome, the possibility for the fern gametophyte to be a different species, rather than L. lineata, could not be eliminated. Monophyly of Lomariopsis is robustly supported, in contrast to previous study by Rouhan et al. (2007).

Fig. 3
figure 3

One of 53 most parsimonious trees (length 243 steps, CI 0.774, RI 0.835, RC 0.646) retained by the parsimony ratchet analysis performed on the trnL-F IGS sequence data. The values above the branches refer to posterior probabilities from Bayesian analysis, whereas those below the branches indicate bootstrap support values. The occurrence of both observed inversions as well as two characteristic indels (5 and 8 nt) are indicated on the tree

Full size image

Interestingly, two hairpin-associated inversions were observed in the spacer approximately 185 nt upstream of trnF, which is different from the previously reported one for bryophytes (Quandt and Stech 2004; Quandt et al. 2004). Inversion 1 (inv 1) is homoplastic and occurs twice in: (1) L. recurvata, L. vestita, L. maxonii, and L. salicifolia as well as (2) L. amydrophlebia and L. wrigthii, whereas inversion 2 unites L. hederacea, L. muriculata, and L. manii. The distribution of both inversions in phylogenetic context is plotted on the tree in Fig. 3.

Morphology of the aquatic fern gametophyte

The thallus of the alleged aquatic gametophyte of Lomariopsis is ribbon-shaped, profusely branched, and one-cell thick throughout, without a midrib or multicellular cushion. Rhizoids are colorless, mostly borne as marginal clusters. It grows indeterminately with active meristematic cells at the rounded apex (Fig. 1c). There are no gemmae, although small lateral branches sometimes detach from the thallus and develop as new individuals. Archegonia and three-celled antheridia are sparsely formed. These characters were also observed in the gametophytes of Lomariopsis spectabilis found in Taiwan (Fig. 1d), although they do not exactly match the strap-shaped Lomariopsis gametophytes described and illustrated by Atkinson (1973).

No associated sporophyte of the Süßwassertang under study has ever been observed. Following transplantation of the gametophyte from water to soil, its growth rate was reduced, and the old portions of the thallus began to die. Unlike the gametophyte in water, rhizoids in soil-grown gametophyte were brown, and numerous antheridia formed along the thallus margins. However, sporophytes did not develop under such conditions either.

Discussion

The utility of different markers in identifying the mysterious gametophyte

The DNA-based identification approach used here shares a similar concept with DNA barcoding, yet the latter tries to utilize more or less universal DNA barcodes. Deciding which barcode to be used for plants is still in progress (e.g., Kress et al. 2005; Chase et al. 2005, 2007; Ford et al. 2009; Hollingsworth et al. 2009). Several of the proposed DNA barcodes, such as the trnL intron (Taberlet et al. 2006), accD (Ford et al. 2009), and rbcL (Schneider and Schuettpelz 2006; Kress and Erickson 2007), were employed in this study for identifying the mysterious thallus; hence, we believe our results could provide a guideline for the future selection of plant barcodes, especially considering the situation of seedless plants.

The maximum identity and E-values from the BLAST results nicely illustrate that the trnL-F IGS is more suitable for species identification in ferns than rbcL, as sequence similarity for rbcL of Süßwassertang and Lomariopsis spectabilis or L. marginata already reaches 96–98%, while sequence identity of the trnL-F IGS of both species with Süßwassertang is only 89%. In addition, the trnL-F IGS amplicon is only about 600 nt compared to 2.1 kb of rbcL and therefore easier to handle in a barcoding approach. However, as the spacer is missing in some green algae (Quandt et al. 2004) and merely reaches 60 nt in derived mosses (Quandt and Stech 2004), its use is limited.

Although more than 21,000 trnL intron sequences (bryophytes >3000, flowering plants >18,000) are recorded in GenBank, ferns are vastly underrepresented, with 342 records (on 2 February 2008), which rendered the database comparison problematic. No trnL intron sequences of Lomariopsis species or Lomariopsidaceae are recorded in GenBank, which explains why the closest matches were found among members of the Dryopteridaceae (Table 1). However, similar to the coding regions, trnL also placed Süßwassertang within polypod ferns. The reported values from BLAST searches representing sequence divergence indicate that the trnL intron resolves more relatively recent divergences compared to rbcL. Likewise, BLAST searches based on the obtained rps4 (plus rps4–trnS IGS) sequence showed only 86% maximal sequence identity of Süßwassertang with L. marginata compared to 96% found for rbcL, indicating the higher potential of rps4-trnS in barcoding approaches compared to rbcL (Table 1). AccD displayed a slightly higher performance than rbcL, with 96% identity to L. spectabilis (Table 1).

Therefore, if rbcL was to be chosen as the DNA barcode, a two-step approach would be favorable. With the similar concept, Kress and Erickson (2007) proposed a two-locus barcode combining trnH–psbA with rbcL for plants. This combination worked well in filmy ferns (Nitta 2008). However, trnH–psbA is absent in black pine (Wakasugi et al. 1994) and since two copies of trnH–psbA can be found in the Adiantum chloroplast genome (Wolf et al. 2003), a careful investigation should be done on whether multiple copies may or may not mislead species identification in ferns. Regardless of the rather conserved rbcL, Lahaye et al. (2008) proposed matK as the prime plant barcode. However, due to the rapidly evolving nature of ferns’ matK and a lack of universal priming sites, especially at the 5′ end (Kuo et al., unpublished data; Wicke and Quandt, unpublished data), it would be problematic to use matK in ferns. Clearly, a comprehensive survey in seedless plants on the utility of different potential barcodes is urgently needed.

An aquatic gametophyte from an epiphytic sporophyte

Fern gametophytes are well known for their extreme tolerance to environmental stresses, such as winter cold (Sato 1982), light deficiency (Johnson et al. 2000), and desiccation (Watkins et al. 2007). As a result, gametophytes in some cases were able to establish populations in sites that were probably far too extreme for sporophytes by exclusively maintaining the gametophyte generation (Farrar 1967, 1990; Dassler and Farrar 1997; Rumsey et al. 1999). Our discovery of Süßwassertang contributes another extraordinary example. Süßwassertang, originally known as a species of bryophytes, has been used to decorate fish tanks. Based on the results of our study involving DNA markers, we have identified Süßwassertang as gametophytes of Lomariopsis, an exclusively hemi-epiphytic fern clade, and found that these gametophytes have an exceptional capability to thrive in water for years without forming their sporophyte counterpart.