Our sampling of L. natans focused on Saxony (Germany), where the species is found in two pond areas approximately 90 km distant from each other: Niederspree (DE_Ni) and Großenhainer Pflege (DE_Gr; Hanspach, 2007; Hanspach et al, 2016). The occurrences in the Niederspree pond area have been known at least since 1899 and in Großenhainer Pflege since 1840.

Both pond areas contain a number of distinct populations in the respective single ponds ( Table 1 ). Within the pond areas several thousand shoots have been detected, in which high fluctuation rates up to 50% were observed (Hanspach et al., 2016, own observations). However, despite historical records (personal communication L. Runge) no occurrences of Luronium were found at Goldgrubenteiche. The material was collected from the end of September to the beginning of October 2020. In accordance with the collection permit issued by the relevant authorities leaves from 1−28 plants per pound (depending on the size of the subpopulation; minimum distance between samples – 0.8 m) were collected and dried in silica gel. To ensure that the harvested material came from one plant, only a single leaf per plant was harvested. In many cases plants occurred in rather dense patches, from which we sampled only one leaf to avoid collecting clonal plants.

Additionally, leaves from plants of two other localities in Germany (Mecklenburg-Vorpommern: near Rostock; Bavaria: Bad Alexandersbad), as well as several samples from the Czech Republic, Poland, Great Britain, and Norway were investigated ( Table 1 ). Furthermore, leaves from plants cultivated in the Botanical Garden of the TU Dresden were also studied ( Table 1 ).

DNA was isolated from silica gel dried leaf material using innuPREP Plant DNA Kit (Analytic Jena, Germany) following the manufacturer’s instructions except for the elution of DNA, which was eluted in two steps: 1. with 70 µl HPLC grade water, 2. with 30 µl HPLC grade water. The quality of isolates was checked with agarose gel electrophoresis (1%) and DNA quantity was estimated using Qubit Fluorometer (Thermo Fisher Scientific). Extracted DNA was stored at -22°C until further processing.

To investigate genetic diversity within Saxon populations and their relationships to other European populations, AFLP (Vos et al., 1995) was used with minor modifications. One hundred nanograms of isolated DNA were digested with the restriction enzymes PstI and MseI. Restriction and ligation were performed in a single reaction for 10 hours at 37°C. For selective PCR, we followed Schuelke (2000). We first tested 15 primer combinations based on previous AFLP studies on L. natans (Cox et al., 2014; Nielsen et al., 2006) and selected four combinations for further analyses: S7 (PstI-ACG/MseI-CAC), S8 (PstI-ACG/MseI-CCG), S11 (PstI-ACT/MseI-CTC), S12 (PstI-ACT/MseI-GCT). Samples were randomly distributed on 96-well plates to avoid position effects. Sixty-six of 253 samples were analyzed twice for quality control. In addition, five identical samples were repeated on each of the five 96-well plates, and negative controls for restriction/ligation, pre-selective PCR, and selective PCR were included. Automated detection of AFLP fragments was performed by the Senckenberg Biodiversity and Climate Research Center (SBik-F; Frankfurt am Main, Germany) with an ABI 3730 sequencer (ABI Life Technologies, Darmstadt, Germany) using the LIZ-600 size standard (ABI Life Technologies).

Fragment scoring was processed as described in Ley and Hardy (2013). First, scoring was automatically done using PeakScanner 1.0 (Applied Biosystems, ThermoFisher Scientific, Berlin, Germany) on a size range from 100–500 bp, the minimal peak height of 30, and maximal peak width of 1. The program TinyFLP v1.30 (Arthofer, 2010) was then used for a pre-choice of markers. Finally, the software SPAGeDi (Hardy and Vekemans, 2002) was used to estimate the reproducibility of bands by calculating fragment-wise Fst values across repeated samples by combining all combinations within one matrix (see details in Ley and Hardy, 2013). We retained a final data matrix consisting of 151 samples from 22 (sub)populations with 48 fragments with a Fst ≥0.25.

To estimate the genetic diversity within populations (for those with >5 samples) we calculated Expected Heterozygosity (He); Shannon’s Index of Diversity (I), and the Percentage of Polymorphic Loci (%P) with GenAIEx v. 6.5 (Peakall and Smouse, 2012). Population structure was analyzed using GenAIEx by performing Analysis of Molecular Variance (AMOVA) and Principal Coordinate Analysis (PCoA) based on Jaccard distances with the R-Package vegan (Oksanen et al., 2020). In addition, we performed Mantel tests (Mantel, 1967) with GenAIEx to check for a correlation between geographic and genetic distances. Furthermore, Bayesian clustering was conducted with the R-package ParallelStructure (Besnier and Glover, 2013) with 10 iterations for every K from 1 to 10, with a burn-in of 500,000 generations followed by 1,000,000 generations. The best-fitting model was chosen according to the method described by Evanno et al. (2005) using the software Structure Harvester v.0.6.94 (Earl and vonHoldt, 2012). Results were visualized with the program DISTRUCT v.1.1 (Rosenberg, 2003).

The experiment started on October 27, 2020 and ended on March 12, 2021. Plant material for the experiment originated from the population Kleiner Tiergartenteich (DE_Gr_Tier; Table 1 ) and had been cultivated since 2019 in the Botanical Garden of the TU Dresden. The plants used for the experiment were propagated by runners from mother plants and were grown after separation from the mother plants for 7−8 weeks in submerged pots in the outside area of the botanical garden.

Conditions tested in this experiment were selected based on observations from habitat and previous publications (Glück, 1905; Barrat-Segretain and Bornette, 2000; Nielsen et al., 2006). Plants were potted into four types of substrates: clay, sand, mixed, i.e. clay/sand mixture (1:1), and layered, i.e. sand as the top layer and clay as the bottom layer (1:1) to check whether lack of nutrients (especially sand as top layer) will influence root length. Pots were subsequently placed in three different water levels: (1) semi-terrestrial condition (saturated): water up to about 3 cm below the rim of the pot; (2) aquatic condition I: water about 0.5 – 1.0 cm above the rim of the pot; (3) aquatic condition II: water about 7 cm above the rim of the pot. Since seasonal differences in plant growth have been observed previously which are likely related to water temperature (Barrat-Segretain and Bornette, 2000; Nielsen et al., 2006), plants were exposed to two different water temperatures: cold: ~ 5 °C and warm: ~ 14 °C. Four plants each, i.e. biological replications were used for each combination of substrate, water level, and temperature ( Supplementary Figure 1 ), thus the experiment constituted a full-factorial design. Water level (rainwater) was regularly checked and replenished when necessary. The plants were illuminated with assimilation lighting from 7 to 10 a.m. Since not all plants used for the experiments were of the same size, their size classes were recorded in the beginning of the experiment: small (s), medium without floating leaves (m), and medium with floating leaves (wfl). After completion of the experiment roots of each plant were rinsed with water to remove remaining substrate. Then, the length of the longest root and of the longest leaf (floating leaves were excluded) were measured. Typically, roots are cut and weighted, but this procedure could not be performed since the plants had to be kept alive for further ex situ conservation.

The effects of the substrate, water level, and temperature on root and leaf length were analyzed using linear regressions. Additionally, we analyzed the effect of the plant size at the onset of the experiment by including the variable as a covariate. For each response variable, we calculated 27 different models ( Supplementary Tables 1 , 3 ) with the predictor variables in plausible combinations. For model selection, we used Akaike’s information criterion (AIC). The 27 models were ranked via AIC and the model with the lowest AIC and a difference of at least 2 to the next model was then considered as the best-fitting model (Burnham and Anderson, 2002). All analyses were performed using the R environment (R Core Team, 2020).

Genetic diversity of L. natans populations was reasonably high Table 2; (grand means He = 0.25; Shannon-Index = 0.387) and differed only little between populations, with highest values found in the population Raschützteich (DE_Gr_Ras: He = 0.268) and lowest diversity detected in the Norwegian population (NO: He = 0.234).

The Principal Coordinate Analyses including all 151 samples across Europe separated mainly samples from Eastern Europe (Czech Republic = CZ and Poland =PL) and those from Eastern Saxony (pond area Niederspree = DE_Ni) from the remaining samples along the first axis ( Figure 2 ). In the left part of the plot samples from the Central Saxon area Großenhainer Pflege (DE_Gr) were clustered, whereas samples from Northern and Southern Germany (DE_Ro, DE_By), Great Britain (UK), and Norway (NO) were found in a rather intermediate position. Note that samples obtained from ex situ collections of Botanical Gardens clustered according to their geographical origin ( Table 1 ). The PCoA presented in Figure 2 was based on Saxon samples only. Accordingly, samples from Niederspree (Eastern Saxony) were clearly separated along the first axis. We observed no genetic structure among the populations within both pond areas, respectively.

Analyses of Molecular Variance ( Table 3 ) revealed a moderate differentiation (ΦPT = 0.334, P ≥ 0.001) between both Saxon localities (DE_Gr and DE_Ni). Results from Bayesian clustering ( Figure 3 ) with two clusters (K = 2) roughly corresponded to the PCoA analyses of all samples. Accessions from the area Großenhainer Pflege (DE_Gr) were assigned to the green cluster, whereas populations from Niederspree were found in the same cluster containing most of the Czech (CZ), Polish (PL), and Norwegian (NO) samples. Admixture between both clusters was detected for a few samples from various ponds in Central Saxony and for plants originating from Norway (NO) and Wales (UK). Mantel test did not detect a significant correlation between geographic and genetic distances (all populations: R² = 0.0271; p = 0.100, Saxon populations: R² = 0.9974 and p = 0.078).

The best-fitting model for root length (adjusted r-squared 0.65, p < 0.001) included an interaction between temperature and water level ( Supplementary Tables 1 , 2 ). The size of the plants at the beginning of the experiment did not have remarkable effects on root length. Roots were longer in plants grown under high water levels than under low water levels. The effect of temperature on the root length was restricted to plants grown under saturated water conditions: here, warm temperature led to roots three times longer than those of plants grown under cold temperatures ( Figure 4 ).

The best-fitting model for leaf length (adjusted r-squared 0.47, p < 0.001) included an interaction between plant size at the onset of the experiment and water level. According to the best-fitting model ( Supplementary Tables 3 , 4 ), leaf length increased with the water level. Plants of medium size and with already developed floating leaves had the longest leaves, and this effect tended to increase with increasing water depths ( Figure 4 ). However, the selected model did not fit much better than the model with plant size at onset and water level lacking interaction, and the model including plant size at onset, water level, and temperature (Δ AIC of the first three models was <2).

In contrast to our initial hypothesis of low genetic variation within populations, the values of the indices of genetic diversity were quite high (I = 0.367−0.416; Table 2 ), and these results suggest that Saxon populations are not genetically impoverished and that clonal reproduction does not cover the effect of the sexual reproduction. This finding was rather unexpected and contradicts the results from Bartušek (2009), who reported a moderate to very low (sometimes absent) genetic diversity in Czech and Saxon populations but applied possibly less polymorphic isozyme markers compared to AFLPs (Ipek et al., 2003; Mondini et al., 2009). Similarly, Kay et al. (1999) suggested the dominance of clonal propagation in the United Kingdom and Ireland based on isozyme analyses, and also previous studies from Belgium reported considerably low values (I = 0.0996) based on AFLPs (Cox et al., 2014). In addition, Szmeja (2004) assumed that L. natans populations in Poland are mainly represented by multigenerational clones reproducing primarily by fragmentation. However, he observed plants in water reservoirs with deeper water bodies, where cleistogamy occurs more often (Kay et al., 1999). Thus, the high genetic diversity observed in the Saxon populations may be due to marked outcrossing from insect-pollinated chasmogamic flowers above the water surface, which we observed in all ponds studied. Moreover, L. natans produces a substantial seed bank and is characterized by high rates of seed germination (average 51-60%; Nielsen et al., 2006). Lansdown and Wade (2003) assumed the seeds to have a capacity for extended dormancy over many years. Germination might take place preferably in shallow waters or on naked mud in empty ponds. The extreme fluctuations of observed shoots (up to 50%) between years (Hanspach et al., 2016, own observations) can be caused by pronounced clonal growth or by high proportion of regeneration from seeds. A much denser sampling design in consecutive years may help to investigate the different ways of population growth. Moreover, estimators of genetic diversity respond often delayed compared to rapid changes in habitats and thus in strong declines in population size (Reichel et al., 2016; Aavik et al., 2019).

In contrast to our results, low values of genetic diversity based on AFLPs (I = 0.025−0.140) were reported within three species of the closely related genus Baldellia (Alismataceae) in Europe (Arrigo et al., 2011). However, these authors found varying values depending on the geographic origin of the samples with higher values detected at Iberian Peninsula compared to France and Switzerland, which were explained by a recent post-glacial re-colonization to the North. Thus, considering a species’ biogeographic history might also be crucial regarding conservation approaches.

According to our hypothesis of a pronounced genetic structure between Saxon populations, L. natans samples clustered mainly according to their geographic origin, whereby we detected three groups of samples: (1) Czech populations, (2) Saxon populations from Niederspree and (3) Saxon populations from the pond area Großenhainer Pflege, with remaining samples from Germany and Norway in between ( Figure 2 ). This partially coincides with the results reported by Bartuška (2009), who showed that the populations from the Czech Republic, and the two Saxon populations Großenhainer Pflege and Niederspree constituted three separate groups. Remarkably, sites from Saxony, which are only separated by a distance of approximately 90 km appeared to be rather distantly related because samples from the area Großenhainer Pflege formed a separate cluster and those from Niederspree were close to remaining samples from various sites across Europe ( Figure 3 ). Most accessions were assigned with high posterior probabilities to one of the two clusters ( Figure 3 ), thus admixture between both Saxon areas played a minor role. Given the rather low geographic distance between Saxon sites the ΦPT value of 0.334 suggests a moderate level of genetic differentiation between populations (Table 3), which was comparable to the values observed in Belgian populations ranging between 0.226 and 0.455 (Cox et al., 2014). Samples from neighboring ponds within each of the respective Saxon areas appeared to be completely intermingled, thus, no genetic structuring on very short distances was observed (Figure 2). While the lack of genetic structure between sub-populations per population ( Table 1 ) may be an effect of water-mediated dispersal (Kay et al., 1999), it does not explain the situation between the three populations in Großenhainer Pflege (DE_Gr_Tier, DE_Gr_Ras, and DE_Gr_Ser) as these areas are not connected by ditches but rather suggests zoochoric dispersal, e.g. by birds. Although, Mikkelsen (1943) argued that Luronium nuts lack adaptations to bird dispersal, more recent studies reported likely dispersals of 50 to 100 km (Fritz, 1989; Halvorsen and Grostad, 2002; Lansdown and Wade, 2003; Nielsen et al., 2006). Here, bird dispersal would only explain the genetic similarity at the local scale ( Table 1 ) but not between areas, because both areas in Saxony (DE_Gr and DE_Ni) were clearly separated ( Figures 2 , 3 ). However, we detected no significant pattern of isolation by distance (Mantel tests for European populations: R² = 0.0271; P = 0.100, for Saxony R² = 0.9974 and P = 0.078) but due to the restricted availability of plant material our sampling was rather unbalanced.

Remarkably the genetic similarity of Polish (PL) and Czech (CZ) samples ( Figure 2 ) is surprising, given the geographic distance between the populations. We would have rather expected a close relationship of the Polish samples with the German population from Rostock (DE_Ro). In Poland, two big geographically separated occurrences, i.e. Pomerania (current populations; closer to Rostock) and Lower Silesia (historical populations; closer to the Czech Republic) are present (Dajdok and Proćków, 2003; Szmeja, 2004). Our findings are in line with the observation of Nischkowsky and Schube (1908) and Suda et al. (2000), who referred to the Lower Silesian occurrences as part of the Jizera Foothills meta-population, which also include the oldest Czech populations. The Czech populations examined in the present study are about 120 km away from these historical occurrences (Nischkowsky and Schube, 1908) but may represent together with scattered populations from Wielkopolska, Poland (Szmeja, 2007; Szmeja, 2013) remnants of a former widespread distribution in Poland and the Czech Republic. The closer relationship of samples from Niederspree (DE_Ni) to Czech populations rather than to the more western Saxon samples from Großenhainer Pflege (DE_Gr) implies a geographic separation of L. natans between Eastern and Western Europe, whose border might cross Saxony.

Our experiment showed that Luronium plants developed longer roots under semi-terrestrial (saturated) conditions at higher temperature ( Figure 4 ). This is in line with the assumption of Glück (1905) who argued that a more extensive root system of semi-terrestrial forms may be caused by their restricted access to water. The here observed longer roots at higher temperatures fit also with Luronium’s Ellenberg Indicator Value for temperature (T6; Ellenberg et al., 2001) indicating its adaptation to moderate warm to warm temperatures. In addition, we found longer leaves in plants growing at higher water level, which can be explained by the fact that the aquatic form of Luronium has longer leaves than the semi-terrestrial form. In addition, plants which were larger at the onset also produced longer leaves, however, plant size at onset was an uncontrolled factor, and the plants were randomly selected according to size. Thus, the best results for obtaining robust plants for potential reintroduction are obtained under semi-terrestrial conditions and at higher temperatures, as a strong root system can facilitate active planting.

Our study revealed moderate levels of genetic diversities in Saxon populations suggesting that the observed decline in populations is not caused by lacking genetic diversity. We assume that populations in ponds rely strongly on regeneration by seeds emerged from outcrossing rather than by clonal reproduction or cleistogamous flowers. While we found a considerable genetic differentiation among Saxon populations our results showed no genetic differentiation within pond systems indicating sufficient genetic exchange between neighboring ponds probably mediated by birds or water. Thus, our results imply to use rather nearby populations from the same area as source for reintroductions. Our plant growth experiments demonstrated that possible reintroductions of plants are facilitated by optimal growing conditions in gardens since we obtained plants with longest roots when grown under saturated water conditions and higher temperatures, whereas the tested soil type seems to have less impact.