A taxonomic revision of
 Lemna
 sect.
 Uninerves
 (Lemnaceae)

Lemna sect. Uninerves Hegelm. consists of three species, Lemna minuta Kunth (synonym L. minuscula), L. valdiviana Phil. and L. yungensis Landolt. Lemna yungensiswas discovered growing on rocks in the Yungas in Bolivia by E. Landolt and was described just 20 years ago. In the original description, Landolt reported that this species is closely related to L. valdiviana and that it is difficult to distinguish the three species on a morphological basis. Therefore, the taxonomic position and status of L. yungensis remained controversial. Here, we carried out a detailed taxonomic study, integrating approaches that include quantitative morphometry, metabolomic profiling by matrix-assisted laser desorption ionization time-of-flight mass spectrometry (MALDI-TOF-MS) as well as molecular genetic analysis using amplified fragment length polymorphism (AFLP), and barcoding of plastidic sequences. We also investigated genome sizes of clones of the three species. Whereas L. minuta can easily be differentiated from L. valdiviana and L. yungensis, it was not possible to distinguishL. valdiviana fromL. yungensiswith any of themethods used. These data imply thatL. yungensis is identical toL. valdiviana. Thus, the name L. yungensis should be synonymised with the name L. valdiviana, since this is the older name.

mainly for ecogeographical reasons, since he found fronds of this species only growing on wet rocks in the province Nor-Yungas, Bolivia (Landolt, 1998), a region belonging to the tropical mountain rain forest, which is rather atypical for members of duckweed. He inserted the species in L. sect. Uninerves mainly based on its high morphological similarity to L. valdiviana. According to Landolt (1998), the main distinguishing features of the two species are the larger size and the more or less symmetric shape of L. yungensis compared to L. valdiviana. Further, papillae along the midline, which are often present in L. valdiviana, are never observed in L. yungensis. Finally, the margin of the frond is sometimes slightly reddish violet in L. yungensis, which is not the case for L. valdiviana. Later, Landolt reported that these markers depend to a large degree on the cultivation conditions and are not very obvious when plants are grown under laboratory conditions (communication of E. Landolt to KJA). The separation of all three species by sequencing of plastidic (Les & al., 2002) or even nuclear DNA fragments (Tippery & al., 2015) has to be considered as not conclusive because only one clone from each species was used. Therefore, we investigated all three species by amplified fragment length polymorphism (AFLP; Bog & al., 2010). In this study, clones of the species L. minuta could be distinguished from those of L. valdiviana and L. yungensis, but L. valdiviana could not be distinguished from L. yungensis. One of the authors, E. Landolt, was, however, concerned that almost all clones of the species L. minuta that were investigated, originated from places were the species was introduced and not native. This might have resulted in a larger genetic similarity between the investigated L. minuta clones than it is the case in nature worldwide. To account for this, L. minuta clones from South America were included in the current study.
Little is known on the reproductive biology of the three species, and the information that can be found throughout the literature is not very satisfying. In general, all duckweed species are assumed to reproduce mainly asexually, though flowering in nature is known for all species with a few exceptions for the genus Wolffia Horkel ex. Schleid. (Landolt, 1986;Les & al., 1997). The proportion of fruiting samples from herbarium specimens and personal observations by Landolt (1986) were ≤2% for L. valdiviana and L. minuta. It is not known, whether hybridisation occurs between the three species.
In the present project, after completing a careful quantitative morphological analysis (cf. Ceschin & al., 2016b), we used as many independent molecular taxonomic methods as possible to test whether L. minuta, L. valdiviana and L. yungensis represent three distinct species within L. sect. Uninerves. Besides fingerprinting by AFLP and barcoding by sequences of plastidic fragments, matrix-assisted laser desorption ionization time-of-flight mass spectrometry (MALDI-TOF-MS, see Materials and Methods) was applied to these three species. Moreover, the genome size was measured to find additional arguments for or against the species delineation.

■ MATERIALS AND METHODS
Cultivation of plants. -All clones were taken from the stock collection of the University of Jena, Matthias Schleiden Institute -Plant Physiology, and are available as living material from the duckweed stock collections in Jena and at Rutgers State University of New Jersey (New Brunswick, New Jersey, U.S.A.) under the international four-digit code (Zhao & al., 2012) given in Table 1. The clones were cultivated in N-medium (Appenroth & al., 1996) under controlled climatic conditions of 25 C and continuous light at 100 μmolÁm −2 Ás −1 PAR, if not stated otherwise.
For molecular genetic analysis, the plant material was separated by filtration from the nutrient medium after ca. 10 days of cultivation, dried on paper towels, snap frozen in liquid nitrogen and then stored at −70 C until use.
To prepare plant material for MALDI-TOF-MS analysis, plants were conditioned to the N-medium for four weeks (precultivation phase) in order to ensure reproducible results. The nutrient medium was replenished every week to prevent nutrient limitation. Details and the further procedures were described by Ziegler & al. (2015). Thereafter, the plant material (1 g fresh weight) was snap frozen in liquid nitrogen and lyophilised.
For morphometric analysis, four clones per species were cultivated for three weeks at 26 C. Again, the N-medium was replaced every week to prevent nutrient limitation.
Morphometric analysis. -Morphological characters were investigated on 25 fully grown mother fronds per clone. In a first step, the presence of papillae was checked. Then the fronds were bleached in NaOCl (0.28%) for 15 to 30 minutes and subsequently washed in distilled water. For better visualisation of the aerenchymatic part and the nerve, fronds were stained with methanolic carmin according to Schwarz (Schömmer, 1949) for 1.5 days. Subsequently, fronds were washed in distilled water and differentiated in acidic isopropanol (50% isopropanol, 0.5% HCl) for 6 to 8 hours. Preparations were immediately photographed using a Leica DM2500 microscope (Leica Microsystems, Wetzlar, Germany) (Fig. 1). The following frond characters were measured using ImageJ v.1.52a (Schneider & al., 2012): (1) length from base to apex (L), (2) width at the widest part (W) of the fronds, (3) length of nerve from node onwards (LN), (4) distance between node and end of aerenchyma (A) and (5) distance from node to apex (N-A) (see Fig. 1D for schematic overview). We also calculated ratios between some of these characters, e.g., to describe whether or not the nerve reaches beyond the aerenchymatic part, which is a diagnostic character according to Landolt (1980) (Table 2). The averages of all characters were tested for significant differences using the Holm-Sidak test in the programme SigmaPlot v.11.0 (Systat Software, San Jose, California, U.S.A.).
AFLP and sequencing of plastidic fragments. -DNA extraction, AFLP fingerprinting and barcoding were done following protocols as given by Bog & al. (2015) unless otherwise noted below. For scoring of AFLP fragments all samples, including 14 duplicates, and 216 parameter combinations for automated analysis of electropherograms (GelCompar II v.5.1; Applied Maths, Austin, Texas, U.S.A.) were screened as proposed by Holland & al. (2008). This resulted in an optimal parameter setting of minimum profiling set to 1.5, minimum area set to 0.3, and matching tolerance set to 0.14. After scoring, only one sample of the 14 duplicates was selected randomly for final analysis. Banding patterns were compiled into a 0/1 matrix (suppl. The methods applied for identification are indicated by +. For AFLP, the 0/1 matrix is given in the supplementary material (Table S1), the accession numbers of the barcoding fragments psbK-psbI and atpF-atpH are given in this  (Paradis & al., 2004), we conducted a principal coordinate analysis (PCoA) based on Bray-Curtis distances to investigate the structure of the dataset. For barcoding, the markers atpF-atpH and psbK-psbI (Wang & al., 2010;Borisjuk & al., 2015) were used. The annealing temperature for the PCR amplification of both plastidic markers was 55 C. All sequencing steps after purification of PCR products were done by Macrogen (Amsterdam, Netherlands). Accession numbers of the sequenced fragments are given in Table 1. The alignment (suppl. Appendices S1 and S2) of the two markers was edited manually using BioEdit v.7.2.5 (Hall, 1999). Indels of the concatenated cpDNA sequences were coded using SeqState v.1.4.1 (Müller, 2005) according to the method of Simmons & Ochoterena (2000). Subsequently, TCS v.1.21 (Clement & al., 2000) was used to build a haplotype network from which we selected unique haplotypes to run jModeltest v.2.1.10 (Darriba & al., 2012). The sequences and indel-coded partitions were tested independently to select the best-fit models of nucleotide substitution. For phylogenetic tree reconstruction, we used the programme BEAST v.1.10.4 . The xml input file was created using the BEAST companion programme BEAUti v.1.10.4 with the following parameters: substitution model for sequence partition -GTR + I with estimated base frequencies; substitution model for indel partition -HKY with estimated base frequencies; molecular clock modelstrict clock; tree prior -Yule process; parameter priorsdefault. The length of the MCMC was set to 10 million steps with sampling of every 1000th step leading to 10,001 trees. The analysis was run three independent times. All three runs were examined for congruence and effective sampling sizes of ESS >200 using TRACER v.1.7.1 . Finally, the runs were processed with two further BEAST companion programmes. They were combined using LogCombiner v.1.10.4 with the first 10% discarded as burnin, respectively, and the tree information was summarised onto a single target tree using TreeAnnotator v.1.10.4 with the following parameters: burn-in = 0, posterior probability limit = 0.5, target tree type = maximum clade credibility, node heights = median heights. As a second phylogenetic tree inference, we used the maximum likelihood (ML) optimality criterion as implemented in the programme RAxML-NG v.0.9.0 (Kozlov & al., 2019). Since RAxML-NG did not run with the same partition scheme as we used for BEAST, we ran the sequence and indel partitions as unpartitioned data under the GTR + I substitution model. We started the ML tree search with 50 random +50 parsimony-based trees and ran D, Overview of the measured characters for the morphological analysis. Fronds were prepared using a methanolic carmin staining method. L = length from base to apex, W = width at the widest part of the fronds, LN = length of nerve from node onwards, A = distance between node and end of aerenchyma, N-A = distance from node to apex. Scale bar represents 1 mm. To estimate the number of species that can be found on the basis of the concatenated and indel-coded cpDNA sequence data, we used two popular methods for single-locus species delimitation. First, we used the Generalised Mixed Yule Coalescent (GMYC) model (Pons & al., 2006) as implemented in the R package splits v.1.0-19/r52 (Ezard & al., 2017). Second, we used the multi-rate Poisson Tree Process (mPTP) method (Kapli & al., 2017), which is accessible via a web service at https://mptp.h-its.org/. GMYC requires as input file an ultrametric tree like the one that was reconstructed by BEAST. The application mPTP is less restrictive and was run with both the tree inferred by BEAST and the tree with the best ML score from RAxML-NG. In both species delimitation calculations, the two outgroup species were discarded.
MALDI-TOF-MS. -Over the last decade, matrix-assisted laser desorption ionisation time-of-flight mass spectrometry (MALDI-TOF-MS) has emerged as a powerful tool for the species-level identification of bacteria (Mellmann & al., 2008), archaea (Krader & Emerson, 2004), yeasts (Marklein & al., 2009) and filamentous fungi (Bader, 2013). The method enables the highly reproducible detection of protein mass patterns (proteomic profiles) that are obtained by analysing whole cells, cell lysates or crude cell extracts. The majority of the signals in the MALDI-TOF mass spectra can be assigned to ribosomal and other highly abundant proteins. These protein patterns have shown to be highly conserved and consistently expressed under a range of growth conditions (Meyer & al., 2017). Our own preliminary tests have shown that species of duckweed can also be successfully differentiated by this method. For proteomic profiling of species of the genus Lemna, MALDI-TOF-MS was carried out with the MALDI Biotyper (Bruker Daltonik, Bremen, Germany) using freeze-dried plant material as described by Blättel & al. (2013). MALDI-TOF-MS was performed on a Microflex LT spectrometer using the software flexControl v.3.0 (Bruker Daltonik). Main Spectrum Profiles (MSP) of each of the clones were created from 20 to 24 single mass spectra.
To check the quality of the mass spectra the flexAnalysis software v.3.3 (Bruker Daltonik) was used. Mass spectra with anomalies were excluded. To evaluate the similarity of the clones investigated, a score-oriented dendrogram was created using the dendrogram creation standard method of the Biotyper v.3.1 software (Bruker Daltonik) which calculates the distance correlation of the different MSP by the average group linkage algorithm.
Flow cytometric genome size estimation. -Genome size measurements were performed using a CyFlow Space flow cytometer (Sysmex Partec, Görlitz, Germany) according to Doležel & al. (2007). Appropriate amounts of fresh fronds were chopped with a sharp razor blade together with corresponding amounts of young leaf material of Raphanus sativus L. convar. sativus, accession "Voran" (IPK gene bank accession number RA 34; 2C = 1.11 pg; Schmidt-Lebuhn & al., 2010) or Glycine max (L.) Merr. convar. max var. max, accession "Cina 5202" (IPK gene bank accession number SOJA 392; 2C = 2.23 pg; Borchert & al., 2007) as internal reference standard. For nuclei isolation, the DNA staining kit "CyStain PI Absolute P" (Sysmex Partec, Görlitz, Germany) was used. Usually, 10,000 events per sample were analysed, and at least four independent measurements per clone were performed. The absolute DNA contents (pg/2C) were calculated based on the mean values of the G1 peak and the corresponding genome sizes (Mbp/1C) according to Doležel & al. (2003). Data were evaluated statistically using the programme R v.3.3.3 (R Development Core Team, 2017). We first checked for normality using the Shapiro-Wilk test (Shapiro & Wilk, 1965). Then we performed one-way ANOVA accounting for unequal variances as implemented in the package userfriendlyscience v.0.7.0 (Peters, 2017), since the integrated Levene test showed heteroscedasticity. Data were further investigated using the pairwise Games-Howell posthoc test. Subsequently, species that showed no significant differences in genome size were combined and tested against the remaining species. Since this dataset did not meet the assumptions of normality and homoscedasticity, a Mann-Whitney U test was used to check whether the two groups exhibit stochastic equality.

■ RESULTS
Morphometric analyses. -Quantitative morphometric analyses (Table 2) were carried out based on characters used to distinguish species by Landolt (1980Landolt ( , 1998. The parameters used are schematically shown in Fig. 1D. The parameters W and LN as well as the ratios LN: A and LN: N-A were significantly different for L. minuta compared to the other two species, L. valdiviana and L. yungensis. In no case were significant differences observed between L. valdiviana and L. yungensis. No other parameters revealed any significance between species. Also, the reported differences in the symmetry of the base of the fronds (Landolt, 1980(Landolt, , 1998 were not evident between the species. When the datasets for L. valdiviana and L. yungensis were combined and tested against that Version of Record of L. minuta, differences in L, A and N-A became significant in addition to those mentioned above. Amplified fragment length polymorphisms. -Clones of L. sect. Uninerves were analysed by AFLP, and the data were evaluated by PCoA (Fig. 2). Thirteen clones of L. minuta, nine clones of L. valdiviana and four clones of L. yungensis were investigated (Table 1). All clones of L. minuta were clearly separated from all clones of L. valdiviana/L. yungensis by principal coordinate axis 1, which explains 31.8% of the total variation. Clones of L. valdiviana and L. yungensis were distributed along axis 2 (14.9%), but no clear separation between the two species could be seen.
Barcoding by plastidic sequences. -The concatenated cpDNA sequence alignment of psbK-psbI and atpF-atpH consisted of 1162 bp and additionally 32 recoded indel sites. All 28 investigated clones, including two outgroup specimens, comprised in total 13 unique haplotypes. Figure 3 shows the maximum clade credibility tree of the Bayesian inference by BEAST. The three L. minuta haplotypes formed one cluster with high posterior probability (PP = 1). A second cluster is comprised of a mixture of five L. valdiviana and L. yungensis haplotypes with a high support value (PP = 0.98). Nevertheless, L. valdiviana represents a paraphyletic group, because three haplotypes build a separate cluster with L. minuta, though with low support (PP = 0.59). The bootstrap scores (BS) of the ML analysis are low to moderately high (BS < 75), and the tree topology is very similar compared to the Bayesian tree. Running the two species delimitation programmes GMYC and mPTP resulted in one proposed species for either programme. There was also no difference in which tree was used as input data for mPTP. These results suggest that the chloroplast fragments have no discriminatory power for the species under investigation.
MALDI-TOF-MS. -Specific and reproducible reference spectra could be generated for each of the clones with the MALDI Biotyper. The similarity analysis of the MALDI-TOF MSP of the different Lemna clones showed clearly two clusters (Fig. 4); one included all selected clones of L. minuta (five clones); the other one included all selected clones of L. valdiviana (three clones) and L. yungensis (four clones) together. The existing substructure of the second cluster did not separate clones of L. valdiviana and L. yungensis but rather comprised a mixture of the species.
Genome sizes. -We also tested whether the three species under discussion can be distinguished based on their genome sizes since Landolt (1986) mentioned a correlation between progress in evolution (in the sense of reduction of morphological structures) with increasing genome size (cf. Wang & al., 2011). Four clones of each species were tested by flow cytometry (Table 1) to establish genome size. An additional clone of L. valdiviana (8685) showed a strikingly higher genome size (559 Mbp/1C) than the other clones of the same species and was therefore not included in statistical testing. Genome sizes were measured as follows: L. minuta (mean ± standard deviation) 365 ± 2 Mbp/1C, L. valdiviana 382 ± 13 Mbp/1C, and L. yungensis 372 ± 2 Mbp/1C. One-way ANOVA revealed significant differences between the tested species (F = 4.69, P = 0.04), which were further investigated by the Games-Howell post hoc test. Solely for the comparison of L. minuta with L. yungensis the difference in genome size was significant (t = 4.04, P = 0.016). No significant differences were observed for the other pair-wise comparisons, i.e., for the other two pairs: L. minuta vs. L. valdiviana t = 2.52, P = 0.160 and L. valdiviana vs. L. yungensis t = 1.55, P = 0.383. When the values for L. valdiviana and L. yungensis were combined, and tested against L. minuta, a Mann-Whitney U test showed that the values for the combined group were significantly different from those of L. minuta (W = 0, P = 0.008).

■ DISCUSSION
In a previous paper, Crawford & al. (1996) stressed that it was very difficult to distinguish the sibling species Lemna minuta and L. valdiviana of L. sect. Uninerves based on morphological characters. These authors used an allozyme-based approach, which enabled them to distinguish most of the clones of these two species. While it has been difficult to distinguish L. minuta and L. valdiviana, the more recently discovered L. yungensis has even higher similarity to L. valdiviana (Landolt, 1998). As Ceschin & al. (2016b) demonstrated the power of quantitative morphometry by distinguishing the invasive species L. minuta from the native species L. minor in Italy, our investigations of L. sect. Uninerves were started with quantitative morphometry. Especially the length of the nerve was a useful parameter for our study species. With this method, it was possible to distinguish L. minuta from L. valdiviana and L. yungensis, but no significant differences could be discovered between L. valdiviana and L. yungensis. Basal symmetry of fronds was mentioned as an important character to distinguish L. valdiviana from L. yungensis by Landolt (1998), but this could not be proven 0 100 200 300 400 500 600 700 800 900 1000 by our observations. Representative fronds from both spe-