Genetic variation in lowland and mountain populations of Tofieldia calyculata and their ability to survive within low levels of genetic diversity

Loss of genetic diversity is expected to be a reason behind the decline of populations of many rare species. To what extent this is true for populations at the range periphery remains to be explored. Alpine species with peripheral lowland populations are an ideal but little-known model system to address this issue. We used 17 microsatellite markers to investigate the genetic diversity and structure of populations of Tofieldia calyculata, a common species in central European mountains, but highly endangered in lowlands. We showed that lowland populations have lower genetic diversity than mountain populations and that the two groups of populations are not clearly differentiated genetically. The species probably survived the last glaciation in refugia in the margins of the Alps and the western Carpathians and some lowland populations likely originated by postglacial colonisation. Some lowland populations may be relictual, but our data did not unequivocally confirm this. Low genetic diversity of lowland populations is likely the result of the reduction of population sizes, limited gene flow, and selfing. Based on data from herbarium specimens from extinct lowland populations, within-population genetic diversity has not changed over the last century suggesting that, under suitable habitat conditions, these populations are able to survive with low levels of genetic diversity. This idea is also supported by the presence of large viable extant populations with very low genetic diversity. Comparisons between modern and historic collection also showed that a large proportion of genetic diversity was lost, due mainly to the extinction of whole populations. Our results provided detailed insight into the recent past of the populations of Tofieldia calyculata, but the genetic diversity of the populations before the twentieth century remains unknown due to the poor quality of old DNA from herbaria samples. Overall, the study indicates that despite reduced genetic diversity, the lowland populations harbour some unique alleles and, with the current levels of genetic diversity, have a chance to survive in the long-term, and thus deserve conservation.


Introduction
It is generally assumed that genetic diversity is crucial for the long-term persistence of plant populations (Frankham et al. 2010). In the centre of a distribution range, populations are expected to be more genetically diverse and less differentiated than populations at the range periphery (Eckert et al. 2008;Hardie and Hutchings 2010). If the species is common in the centre of the distribution, the necessity to protect peripheral populations, is controversial (Hunter and Hutchinson 1994;Hardie and Hutchings 2010). Recent studies, however, suggested that peripheral populations may contain high genetic diversity (Wagner et al. 2011;Plenk et al. 2017;Kropf et al. 2020) and can be genetically differentiated from central populations (Reisch et al. 2003;Plenk et al. 2017). These peripheral populations may represent sources of genetic diversity that allow species to cope with changing environmental conditions (Hampe and Petit 2005;Gibson et al. 2009), representing sources of evolutionary innovations (Hardie and Hutchings 2010). Understanding genetic differentiation among central and peripheral populations is thus crucial for setting priorities in species' conservation.
Species common in central European mountains (the Alps and the Carpathians) with peripheral endangered populations in the lowlands of central Europe are potential examples of species that have such strongly-differentiated peripheral and central populations (Reisch et al. 2003;Štěpánková 2010;Duwe et al. 2017). In the lowlands, these species are restricted to scarce primary or secondary open habitats (fens, meadows, dry grasslands and rocks) and are often rare and declining. For setting conservation priorities, it is very important to distinguish if the lowland populations are genetically different and relict, or if they are genetically depauperate and not differentiated due to being a product of postglacial colonisation from glacial refugia situated on the margins of the Alps (Schönswetter et al. 2005) or the Carpathians (Mráz et al. 2007;Ronikier et al. 2008;Kolář et al. 2016).
Genetic diversity of species with alpine-lowland distributions is poorly understood. Only a few recent studies focused on species with this type of distribution (e.g., Greimler and Dobeš 2000;Lutz et al. 2000;Reisch et al. 2003;Windmaißer et al. 2016;Duwe et al. 2017;Duwe et al. 2018;Knotek and Kolář 2018), with mixed results. Genetic diversity of lowland populations is often reduced due to their isolation and small size (Duwe et al. 2017;Duwe et al. 2018). In other species, lowland populations show no genetic erosion and high levels of genetic diversity, probably due to high longevity of individuals and population sizes large enough for long-term preservation of genetic diversity (Lutz et al. 2000;Reisch et al. 2002Reisch et al. , 2003Knotek and Kolář, 2018). In these studies, the relict status of lowland populations was evidenced either by high genetic diversity and presence of rare alleles (Greimler and Dobeš 2000;Knotek and Kolář 2018) or by low within-population diversity but high betweenpopulation differentiation due to their high isolation and bottleneck effects (Reisch et al. 2003).
If low genetic diversity is found in some of the lowland populations, it may have originated by two alternative processes. These populations may have always been depauperated due to being the result of long-distance dispersal after the last ice-age (Wroblewska 2008;Windmaißer et al. 2016). Alternatively, the populations could be relict, and their low genetic diversity may be a result of their recent strong size reduction leading to strong genetic drift. Low within-population diversity but high differentiation of populations is expected in this scenario (Reisch et al. 2003;Hensen et al. 2010;Vogler and Reisch 2013). Distinguishing whether the populations are relict and able to survive with low genetic diversity is important when deciding whether it makes sense to protect these populations or not. Such knowledge can be obtained by exploring genetic structure of populations within the distribution range.
Important knowledge can be also obtained by analysis of herbaria material, which can help to distinguish whether a decline in genetic diversity is rather ancient or if it is rather recent and caused by human-mediated habitat loss. Despite some difficulties related to the quality and fragmentation of old DNA, analysis of old herbarium samples has recently been successfully used in some endangered species to test how much genetic variability was lost in the last decades (Cozzolino et al. 2007;Frey et al. 2017;reviewed by Bieker and Martin 2018).
Using microsatellite markers on samples both from central mountain and peripheral lowland populations, we compared genetic diversity and composition of lowland and mountain populations of Tofieldia calyculata (L.) Wahlenb., a common plant species in the Alps and the Western Carpathians, with strongly declining peripheral populations in lowlands of central and north-eastern Europe (Gabrielová et al. 2011;Kaplan et al. 2015). We also analysed herbarium specimens from historical lowland populations since the nineteenth century. These data should help to suggest optimal conservation actions for the species. We asked the following questions: (1) What is the genetic structure of mountain and lowland populations and what does it tell us about the possible origin of the lowland populations? (2) Can the differences in genetic diversity among populations be explained by their lowland/mountain location and population size? (3) How much genetic variability was lost due to the extinction of some lowland populations and the decline in size of the remaining ones?

Study species
Tofieldia calyculata is a small, perennial plant from the family Tofieldiaceae. The species is diploid (2n = 30, Šmarda 2018). Each individual consists of rosettes with several ensiform leaves. The inflorescence is a raceme with about 40 yellow flowers (Štěpánková 2010). The fruit is a capsule with several dozens of tiny, (0.048 mg), brown seeds. The species has a mixed reproductive system with possibility of selfing (personal observation from natural populations, see also S1 in Supplementary Material) and clonality. Clonal reproduction by rhizomes is, however, limited to the close vicinity of mother plant (Klimešová et al. 2017). The species is distributed mainly in the mountains of central Europe. The Alps and the western Carpathians are the 1 3 centre of the distribution range of the species. Peripheral populations occur in the lowlands of central Europe (the Czech Republic, Poland, Austria) with some isolated localities in the Pyrenees, Romania and north-eastern Europe (the island of Gotland and the Baltic states). In lowlands, the species has disappeared from most of its historical localities, and it is highly endangered (Kaplan et al. 2015;Kopeć and Michalska-Hejduk 2012). It is classified as vulnerable in Poland and in the Ukraine, endangered in Lithuania, critically endangered in the Czech Republic and in Croatia, and extinct in both Latvia and Hungary (Gabrielová et al. 2011).
In the past, the species used to be quite common in several regions of the Czech Republic. Drainage, land-use changes and ongoing succession have, however, caused its disappearance from almost all localities (Kaplan et al. 2015). From more than 50 localities recorded since the nineteenth century, the species currently grows in only five of them (Kaplan et al. 2015). The size of those remaining has been strongly reduced and currently ranges from 1 to 400 individuals (unpublished data).
We collected samples from 30 recent populations of T. calyculata (Table 1 ) covering all parts of the range of the species, including 13 central (mountain) populations and 17 peripheral (lowland) populations. When possible, we collected leaves from at least 20 individuals in every population. Leaves were collected from individuals equally distributed across the population. The individuals sampled were at least 2 m apart. Since clonal reproduction is limited to close vicinity of the mother plants, sampling clones was thus avoided. Only in two small populations (Cikánský dolík and V Bahnách), where all the individuals grow in small area, we did collect samples from plants which were closer than two metres. Leaves were dried and stored in silica gel. The population size was estimated by estimating the number of flowering individuals. We were unable to record the number of vegetative plants as, in dense vegetation, these can be easily overlooked.
We also obtained leaves from 135 individuals from herbarium specimens from six herbaria in the Czech Republic and Switzerland (PRC, PR, CB, HR, LIT, BERN, see Table 2 and Table S2 in Supplementary Material) collected between 1864 and 2011 with populations originating mostly from the Czech Republic, but additionally from Italy, Austria, France, Switzerland, Slovakia and Sweden (Gotland). Herbarium samples were used in STRU CTU RE, PCoA and AMOVA analysis (Fig. 1, for detail, see below). The subset of herbaria samples (samples from the Czech Republic) was also used to explore the loss of genetic diversity in populations of T. calyculata in the Czech Republic since the nineteenth century.

Genetic analysis
We used 19 previously developed microsatellite markers for the analysis (Vlasta et al. 2020). DNA was extracted from leaf samples using a modified CTAB method (Lodhi et al. 1994) with all volumes downscaled 10 ×. DNA amplification, was completed in two multiplex reactions (previously described in Vlasta et al. 2020) containing 2.5 µL of QIAGEN Multiplex PCR Master Mix and 20 ng of DNA dissolved in 0.5 µL of distilled water. For multiplex mix I (MM I), the PCR contained 1.94 µL of primer mix (10 µM each in initial volume) and 0.06 µL of H 2 O; for MM II the PCR contained 1.875 µL of primer mix (10 µM each in initial volume) and 0.125 µL of H 2 O. The following Thermocycling conditions were used: an initial denaturation step at 95 °C for 10 min; followed by 35 cycles of denaturation (95 °C for 30 s), annealing (60 °C for 40 s), and extension (72 °C for 30 s); and a final extension at 72 °C for 8 min. PCR products were diluted with ddH 2 O 10 × (PCR product of MMI) and 20 × (PCR product of MMII). Each PCR product (1 µL) was mixed with 12 µL of formamide combined with 0.1 µL of size standard (GeneScan 500 LIZ; Thermo Fisher Scientific, Waltham, Massachusetts, USA). Fragment lengths were determined by capillary gel electrophoresis with an ABI 3130 Genetic Analyzer using GeneMapper 4.0 (Thermo Fisher Scientific).
For analysis of herbarium specimens, the Thermocycler conditions were modified by prolonging time for denaturation, annealing and extension steps: an initial denaturation step at 95 °C for 10 min; followed by 35 cycles of denaturation (95 °C for 90 s), annealing (60 °C for 120 s), and extension (72 °C for 120 s); and a final extension at 72 °C for 8 min. PCR products were not diluted (for herbarium specimens from 1864 to 1939) or were diluted 5 × (for herbarium specimens from 1940 to 2015). The prolonged times for the denaturation, annealing and extension steps led to higher amplification success.

Data analysis
We excluded 2 loci from further analyses due to a high proportion of missing alleles. All analyses were thus done with 17 loci. We also excluded individuals from herbarium specimens with high proportion of missing data (less than 70%   Table 2 List of populations of T. calyculata from herbaria used in this study, (populations and samples with unsuccessful amplification and herbaria accession numbers not included in the table-see Table S2 in Supplementary

Genetic diversity of lowland and mountain populations
We divided populations into two groups: (i) mountain populations (from the Alps and the Carpathians where the species is common and not endangered, with the altitude ranging between 650 and 2045 m a.s.l.); (ii) lowland populations (from the Czech Republic, Poland, Estonia, Gotland and Austrian lowlands where the species is not common and is endangered, with the altitude ranging between 4 and 585 m a.s.l.).
We used SPAGEDi (Hardy and Vekemans 2002) to calculate the number of alleles, the observed and expected heterozygosity, allelic richness (Ar), F ST and individual inbreeding coefficient (F I ). Additionally, we calculated the percentage of polymorphic loci, the number of genotypes in every population, counted the number of unique alleles for each population (i.e., alleles which are present in only one population) and for lowland and mountain populations (i.e., alleles found only in one of these two groups). GENEPOP version 4.2 (Rousset 2008) was used to test deviation of each population from the Hardy-Weinberg equilibrium. Individuals from herbarium specimens from the same locality collected in different years were analysed as one population if the time between years did not exceed 20 years. Populations with less than five individuals were excluded from these analyses.
We calculated pair-wise correlations among all the genetic characteristics using MS Excel. All except F I were highly correlated to each other (see Table S3 in Supplementary Material). We thus only analysed F I and Ar as representative of all the other variables and commonly-used measures of genetic diversity in similar studies.
We tested the effect of population type (lowland and mountain) and logarithm of population size on genetic characteristics of populations (Ar and F I ) using generalised linear model with Gamma distribution (in case of Ar) and ANOVA (in case of F I ) using R version 3.6.3 (R development team 2017). Populations with less than five individuals were excluded from these analyses. We also excluded all populations from herbaria for this analysis, so we compared only current populations.
We used BOTTLENECK version 1.2.02 (Piry et al. 1999) to detect recent reductions in population size. This is done by testing for deviation from mutation-drift equilibrium, which is reflected by a higher heterozygosity (He) than would be expected given the number of allele and a specified mutation .71* -model. We used the two-phase mutation model with 95% single-step mutations and a variance of 12% that is recommended for microsatellites (Piry et al. 1999) and significance was tested by Wilcoxon signed-rank test. Further, we used the modified Garza-Williamson index (Garza and Williamson 2001) to detect evidence of ancestral population decline. The index was calculated for each population in Arlequin 3.5 (Excoffier and Lischer 2010) as the ratio of the number of alleles to range in allele size. Populations with less than ten individuals were excluded from BOTTLENECK analysis.

Genetic structure of the populations
STRU CTU RE analysis (Pritchard et al. 2000) was performed to reveal the structure of the populations. First, we removed all clones from the analysis using AFLPDAT function clone (Ehrich 2006) in R version 3.6.3 (R development team 2017). All the remaining samples were used, including all samples from herbarium specimens with more than 70% loci amplified (altogether 408 individuals). We excluded historical herbarium samples that duplicated our recent data (i.e., herbarium specimens from the same populations from different time periods). The analysis was run with K from 1 to 10 with a burn-in period of 100,000 iterations and run for 1,000,000. Ten individual runs were performed for each K. The admixture model was used. Optimal K was determined by the Evanno method (Evanno et al. 2005) based on the ΔK statistic using STRU CTU RE Harvester (Earl and von Holdt 2012). We also performed PCoA analysis based on Jaccard coefficients using PAST version 2.17c (Hammer et al. 2001). To test the distribution of genetic variability between lowland and mountain populations and between populations within the groups and within populations, we performed AMOVA using Arlequin (Excoffier and Lischer 2010). Three analyses were carried out-one including two geographic groups (lowland and mountain) and two separate analyses of each group.
Results from STRU CTU RE showed the presence of two genetic clusters (groups) with some mixed populations between them. These two groups were then analysed separately to reveal the sub-structure of these two groups using both STRU CTU RE and PCoA. Mixed populations (i.e., populations < 0.8 probability of belonging to first or second cluster) were not included in these analyses. In the first group, 18 populations with 169 individuals were analysed; in the second group, 34 populations with 186 individuals were analysed.

Analysis of herbarium samples
We used a generalised linear model with Gamma distribution to test the effects of herbarium specimen age and location at which they were stored on the amplification success using linearly-transformed percentage of amplified loci as a dependent variable. We were not able to test the interaction between the two variables as plants from different herbaria originated from different years. In this analysis, we used all the data from herbaria. R version 3.6.3 (R development team 2017) was used for this analysis.
In the next analysis, we used only samples from the Czech Republic, both current samples and herbaria historical samples, from 1864 to 1986. Samples were divided into six groups based on the geographic distribution of the species (Table 2). In four regions (Džbán, Polabí, Českolipsko region and southern Bohemia), where current populations are present, for each region we counted all alleles amplified for all loci in all samples (even in samples where less than 70% loci were amplified). We compared the alleles from the past with alleles in the current populations. We divided all alleles found in each of the four regions into three groups: (1) alleles present in current populations; (2) alleles which have disappeared from current populations (i.e., present in the historical, but not recent populations, genetic loss at the within-population level); (3) alleles which have disappeared from extinct populations (i.e., present in historical populations and lost because these populations no longer, genetic loss at the between-population level) and estimated the proportion of alleles in each of the groups.

Genetic diversity of lowland and mountain populations
We found 158 alleles in the 17 scored loci. The number of genotypes ranged between 1 and 20 genotypes in a population (Tables 1 and 2). The percentage of polymorphic loci in a population ranged between 0% and 100%. Ar ranged between 1 and 2.09 (Tables 1 and 2; Fig. 2a). The observed and expected heterozygosity ranged between 0 and 0.432 and 0 to 0.512, respectively (Tables 1 and 2). The highest genetic diversity was found in the mountain population Weißenbach in the eastern Alps. The lowest genetic diversity was found in three lowland populations: Ojcowski Park Narodowy in Poland and at Cikánský dolík and V Bahnách in the Czech Republic. These three populations each contained only one genotype. The highest number of unique alleles was found in population Weißenbach (seven unique alleles). Other unique alleles were found in other mountain populations, whereas few unique alleles were found in lowland populations (see Fig. 2; Tables 1 and 2). Fifty alleles (31.6%) were found only in mountain populations, whereas 12 alleles (7.6%) were found only in lowland populations. We found the presence of null alleles in some populations and loci, but the frequency of null alleles was generally very low. 80% of loci and population had lower than 5% of null alleles (FreeNA).
Lowland populations had reduced Ar compared to mountain populations (Table 3; Figs. 2 and 3). Individual inbreeding coefficient (FI) did not differ between lowland and mountain populations (Table 3; Fig. 3). Neither Ar, nor FI were influenced by population size (Table 3).
We found significant deviations from Hardy-Weinberg equilibrium in almost all populations (Genepop, p < 0.05) except two mountain (Le Taillefer, Mount Rax) and two lowland populations (Bagno Serebryskie, Jestřebské slatě). Inbreeding coefficients were also highly positive and significantly different from 0 in almost all populations, with the exceptions of populations Bagno Serebryskie and Jestřebské slatě (Tables 1 and 2). Values of the modified Garza Williamson Index were low in all studied populations (ranging between 0.15 and 0.46, Table 1). We found evidence of a genetic bottleneck in some lowland populations (Jestřebské slatiny, Sosnowiec-bory, Jaworzno Szczakowa, Ojcowski park narodowy) and in one mountain population indicating recent reduction in population size in some populations (BOTTLENECK, p < 0.05).

Genetic structure of the populations and origin of lowland populations
AMOVA analysis revealed that 7.25% of variability was between lowland and mountain populations and 54.2% of variability between populations within groups and 38.55% within populations. In separate analyses of each group, 44.7% of variability was between populations in mountains whereas 68.2% of the variability was between populations in lowlands (Table 4).
In STRU CTU RE analysis, the most likely number of clusters was assigned to two groups based on the ΔK statistic (K = 2, Tables S4 and S5 in Supplementary Material, Fig. 1a, b). The first group consists of populations from the eastern Alps, the western Carpathians, Estonia and Gotland. The second group contained the remaining populations -from northern Bohemia, Poland, the central, southern and western Alps and the Pyrenees. Populations from the lowlands of Austria and from southern Bohemia were mixed. These results were also supported by PCoA analysis (Fig. 4). Lowland and mountain populations were not distinguished.
After separate STRU CTU RE and PCoA analyses, three sub-groups were found within the first group (K = 3, Tables S4 and S5 in Supplementary Material, Fig. 1c), dividing populations from (1) Estonia and Gotland, (2) the Carpathian populations and (3) populations from the eastern Alps (Figs. 1c and 4). Within the second group, two subgroups were recognised after a separate STRU CTU RE analysis (K = 2, Tables S4 and S5, Fig. 1c); populations from the central, southern, and western Alps together with populations from southern Poland formed one sub-group, whereas populations from northern Bohemia and south-eastern Poland (Lublin region) formed another group (Figs. 1c  and 4).

Recent changes in genetic diversity in lowland populations from the Czech Republic
The percentage of amplified loci was significantly affected by the age of a herbarium specimen (GLM, Explained deviance = 40.83, p < 0.001, Fig. 5), and also by their place of storage (GLM, Explained deviance = 36.06, p < 0.001). Analysis of herbarium specimens from the Czech Republic showed that 43.4% of alleles observed in herbarium specimens was lost due to extinction from almost all historical localities. Comparison of the four main regions where the species was present in the past shows that more genetic diversity was lost in some regions than in others (Fig. 6). In the Polabí region and in southern Bohemia (where only a few individuals are currently present) around half of the alleles was lost (59%, and 49% respectively). In the Českolipsko and Džbán regions, where current populations with hundreds or dozens of individuals still exist, only 21% and 14% of variability was lost respectively. Almost all Czech historical populations had a low within-population genetic diversity compared to central mountain populations, and the populations were highly differentiated. The majority of the genetic variability was therefore lost due to extinction of whole populations.

Genetic diversity of lowland and mountain populations
We found significantly lower genetic diversity in lowland peripheral populations compared to mountain populations. Low genetic diversity (Vogler and Reisch 2013;Duwe et al. 2017) and high genetic differentiation (Greimler and Dobeš 2000;Vogler and Reisch 2013;Duwe et al. 2017) were also found in other species with remnant lowland populations, although other studies showed extremely low genetic diversity in lowland populations compared to the mountains Fig. 1 a and b Two main genetic groups of populations of Tofieldia calyculata based on STRU CTU RE (k = 2), c Substructure of the two main genetic groups depicted in Fig. 1a and b based on a separate STRU CTU RE analysis of the two main groups. The first (blue) group was optimally divided into three (K = 3), the second (green) group into two clusters (K = 2). The black line indicates the border between mountain and lowland populations ◂ 1 3 (e.g., Machon et al. 2001;Kirschner et al. 2011). Our results indicate that despite reduced genetic diversity of lowland populations, they still contain a relatively high amount of genetic variability, including some unique alleles and alleles which are not present in mountain populations. This is consistent with results from other studies on species with similar distribution (Reisch et al. 2002;Reisch et al. 2003;Duwe et al. 2018) reporting more genetic variability in mountains, but with sufficient genetic variability in lowlands.
We did not find any relationship between genetic diversity and population size. A positive relationship between population size and genetic diversity is well-supported in the literature (Tomimatsu and Ohara 2003;Leimu et al. 2006;Šmídová et al. 2011;Dostálek et al. 2014;Münzbergová et al. 2018), although other studies showed no relationship (Pluess and Stöcklin 2004;Bachmann and Hensen 2007;Klank et al. 2012). While some lowland populations of T. calyculata with very low genetic diversity are very small, we also found no genetic diversity in a large viable (based on our unpublished demographic data) population in Ojcowski Park Narodowy, where all the plants sampled had the same genotype. We also found very low genetic diversity in one large Carpathian mountain population (Pusté Pole). Similarly, almost no genetic diversity was found in the Estonian populations. Estonian populations are located on the island of Saarema and their low genetic diversity could be a consequence of postglacial colonisation of this formerlyglaciated area. The low genetic diversity in several large viable populations suggests that low genetic diversity may not be a problem for the species. Although there are many studies that showed the importance of genetic diversity for fitness of individuals and plant populations (e.g., Luijten et al. 2002;Picó et al. 2007;Ilves et al. 2013), there is evidence from other rare species which are able to survive and form viable populations even with low levels of genetic diversity (Lammi et al. 1999;Noel et al. 2007;Vogler and Reisch 2013;Vasquéz et al. 2016;Plenk et al. 2019). It is thus likely that loss of genetic diversity is not the key reason for the decline of the populations of T. calyculata. This decline is probably caused mainly by the loss of suitable habitat conditions caused by drainage, or cessation of traditional management (Štěpánková 2010(Štěpánková , Kaplan 2015. The low values of the modified Garza Williamson index indicate that both lowland and mountain populations of T. calyculata experienced a reduction of population size in the past, suggesting that the reduction of population size and genetic diversity is common in the populations of the species. T. calyculata is a long-lived perennial species, so it is possible that negative consequences of loss of genetic diversity will appear in future generations (but see the section below about herbaria samples). Further, we did not study the effect of genetic diversity on the fitness of the individuals.
What are the reasons for the low genetic diversity of lowland populations? T. calyculata is self-compatible (field observations, also see S1 in Supplementary Material). This is in line with significant deviance from Hardy-Weinberg equilibrium in almost all populations and high F I . Selfing species have lower genetic diversity and higher differentiation of population (Hamrick and Godt 1996;Vasquéz et al. 2016) and they are less affected by low genetic diversity than self-incompatible species (Hamrick and Godt 1996;Honnay and Jacquemyn 2007;Segarra-Moragues and Mateu-Andrés 2007). F I did not differ between lowland and mountain populations indicating that selfing probably happens in all populations. Selfing of the species is also supported by the results of the AMOVA analysis, which revealed a higher proportion of variability between than within populations, with the difference being larger in lowland populations. This indicates higher selfing and isolation of lowland populations, possibly explaining their lower genetic diversity. Another possible explanation of low genetic diversity of some population can be a high proportion of clones. T. calyculata can clonally reproduce by short rhizomes. This type of reproduction is, however, limited (Klimešová et al. 2017). Some populations were formed by only one or few genotypes. The low number of genotypes is rather a result of selfing, along with reductions in population size and restricted geneflow than a result of clonal reproduction, because sampling clones was minimised by collecting samples from plants at least 2 m apart. Individuals in some populations which were more than 100 m apart had the same genotype. Another explanation for the low genetic diversity found in some populations is reductions in population size in the past. This was supported both by BOOTLENECK analysis and low values of modified Garza Williamson index. The low within-population genetic diversity cannot be caused by null alleles, because the frequency of null alleles was quite low across populations and loci.

Genetic structure of the populations and origin of lowland populations
Results of the STRU CTU RE analysis did not confirm the clear separation of lowland and mountain populations and showed that the situation is more complex. Both lowland and mountain populations form several independent groups. Populations from the limestone north-eastern Alps are more closely related to Carpathian populations than to the rest of the populations from the Alps. A similar genetic split between populations from the western and eastern Alps was described in a range of previous studies (Ronikier et al. 2008;Thiel-Egenter et al. 2011;Ronikier et al. 2012;Slovák et al. 2012). While the eastern Alps and the western Carpathians were not glaciated or hosted only small glacial sheets, the central and western Alps were covered by a huge glacier (Ivy-ochs 2015). The eastern Alps and the western Carpathians are therefore proposed as glacial refugia for many cold-adapted species ( Schönswetter et al. 2005;Ronikier et al. 2008;Slovák et al. 2012;Dítě et al. 2018). In line with this, the highest number of unique alleles and the highest values of Ar and heterozygosity were found in populations on the southern and north-eastern edges of the Alps and in some western Carpathian populations, suggesting possible glacial refugia for T. calyculata. Suitable habitats including calcareous fens were probably present during the last ice-age in non-glaciated parts of the Alps and the western Carpathians (Janská et al. 2017). After tree expansion in the Holocene, fen species such as T. calyculata likely survived on old fens originated in the late glacial or at the beginning of the Holocene (Hájek et al. 2011;Hájková et al. 2020). Indeed, the highest genetic diversity within the western Carpathians was found in the population in Belianské lúky, the largest Slovakian fen, more than 10,000 years old and hosting many relict species (Hájková et al. 2012). Populations from the western Carpathians or the eastern Alps could be a source of postglacial colonisation of northern areas (i.e., Estonia and Gotland) as shown by the high similarity of these populations in our study. The role of the Carpathians as the source of postglacial colonisation of the Baltic area was recently also described in Arabidopsis arenosa (Kolář et al. 2016).    The presence of glacial refugia in lowlands cannotbe excluded. Lowland populations have low within-population but high between-population genetic diversity. This pattern was previously assigned to relict populations (Reisch et al. 2003;Hensen et al. 2010). While in mountain populations there was 44.2% variability among populations, in lowland populations 68.2% of variation was among the populations, which indicates restricted gene flow among lowland populations. Populations of T. calyculata in lowlands could have already existed during the last glacial period. Their low genetic diversity could be due to recent reduction of population sizes together with restricted gene flow leading to strong genetic drift during the Holocene. Reduction of population size of some The bars indicate how many alleles survived in current populations, how many alleles have been lost on within-population level and how many alleles have been lost on between-population level (due to extinction of whole populations) populations in the past was supported by a BOTTLENECK analysis and further by low values of the Garza Williamson Index which were below the value proposed by Garza and Williamson (2001) as critical (i.e., 0.68) in all populations. Differentiation of populations then probably arose through selfing.
Possible lowland occurrences of T. calyculata during last ice-age are also supported by palynological records. Pollen of Tofieldia sp. was found in late Glacial or at beginning of Holocene in Central Bohemia (Pokorný et al. 2015), in Estonia (Niinemets et al. 2002) and in the lowlands of Germany (Brande 1996) and in Poland (Ralska-Jasiewiczowa and Rzętkowska 1987). Pollen of T. calyculata cannot be distinguished from pollen of Tofieldia pusilla, which currently co-occurs with T. calyculata in the Alps and the Tatra Mountains, but the distribution of the species does not overlap in lowlands nowadays. It cannot, however, be excluded that pollen found in the palynological records was pollen of T. pusilla.
The relict character of the lowland populations is further supported by the presence of unique alleles in some lowland populations. Unique alleles were found in lowland populations in the Czech Republic, Poland, Estonia and in populations from the lowlands of Austria. In the Czech Republic, we found some unique alleles in two historical populations (Polabská černava and Cibulka). In the region of Polabská černava (Polabí region), the species grew in old fen localities whose history goes back to the Late-Glacial period (Petr and Novák 2014). The species cooccurred with other fen species considered as relicts such as Cladium mariscus (Pokorny et al. 2010) or Ligularia sibirica (Hendrych 2003;Šmídová et al. 2011).
An alternative hypothesis for the origin of lowland populations of T. calyculata can be postglacial colonisation from glacial refugia, as previously described in Primula farinosa (Theodoridis et al. 2017), the species with quite similar distribution and ecology, or in Polygala chamaebuxus (Windmaißer et al. 2016). This hypothesis does not contradict the arguments mentioned above. In this scenario, lowland populations could be considered as postglacial relicts with sufficiently long history to allow occurrence of some new genetic variability. This hypothesis cannot be excluded, especially owing to the lower genetic diversity of lowland populations. Postglacial expansion is probable, especially in populations from north-eastern Europe (e.g., Estonia, Gotland). To conclude, lowland populations probably arose from both processes-postglacial colonisation from glacial refugia and interglacial contraction of relict lowland populations. To examine the origin of lowland populations, it would be appropriate to perform another analysis with more robust data, for example analysis based on Approximate Bayesian Computation approach (ABC, Beaumont 2010).

Recent changes in genetic diversity in lowland populations in the Czech Republic
When performing genetic analysis of herbarium samples, there are still some methodological problems such as different sampling of plants, missing data, mistakes in labelling and a lower quality of DNA due to the age of specimens and the conditions of their storage (recently reviewed by Bieker and Martin 2018). We found that a percentage of amplified loci was significantly affected by the age of the herbarium specimens with lower amplification in older samples, though some old herbarium specimens from nineteenth century were successfully amplified. Using herbarium samples for testing loss of genetic diversity in rare species is still not common (but see Cozzolino et al. 2007;Beatty et al. 2014;Frey et al. 2017). Cozzolino et al. (2007) documented loss of genetic diversity and increased differentiation of populations of Anacamptis palustris after human-mediated landscape changes. Shifts in genetic diversity were also documented using herbarium specimens of Trapa natans (Frey et al. 2017) and Saxifraga hirculus (Beatty et al. 2014). In T. calyculata, even when corrected for the different sample sizes of current and historical populations (Ar values), historical populations had very low within-population genetic diversity and high among-population diversity. Comparison of a pool of alleles from herbarium specimens with a pool of alleles from present populations suggests that 43.4% of alleles observed in the herbarium specimens were not present in current populations, and they were lost mainly due to the extinction of whole populations (Fig. 5). In Polabí and in southern Bohemia regions, the loss was large (59% and 49% of variability lost respectively), with only a few plants now remaining. By contrast, in regions still hosting remaining populations (i.e., Českolipsko and Džbán) only 21% and 14% of the variation (respectively) was lost. In the largest Czech population-Jestřebské slatě (Českolipsko region), the same alleles were present in the 1930s, the 1980s and today. Some other alleles were found in this population in 1910s, but the low percentage of amplified loci did not enable us to draw conclusions about the genetic diversity before 1930s. In the second largest population, Cikánský dolík (Džbán region), all plants have the same genotype. In this population, we found a few additional alleles in the herbarium specimens from the 19th century, but within-population genetic diversity was already negligible in the nineteenth century.
Low within-population genetic diversity identified in the historical populations must be interpreted with caution. In some cases, it could originate from PCR amplification errors, since, in many samples, some loci were not amplified. The percentage of amplified loci was especially low in herbarium samples originating before 1930s. Since we obtained sufficient reliable data from many populations from samples collected after 1930s, our research does, however, 1 3 provide a sound insight into the recent past of the populations of T. calyculata in the Czech Republic. The changes in genetic diversity could have happened before the start of the twentieth century. The BOOTLENECK analysis showed the recent reduction in population size in some lowland populations including some populations from the Czech Republic. The low number of samples from the 19th century and the poor quality of their DNA do not allow us to assess the full diversity of the populations from this period. It is also true that the number of herbaria samples we used is not as big as would be ideal. Further, it is not as big as in the more recent populations. We collected samples from all large Czech herbaria, but the number of specimens in these herbaria is limited and it was not possible to obtain more samples from historical Czech populations. It is possible that using more historical samples could indicate higher within-population genetic diversity in the past. On the other hand, we documented the loss of many alleles, which are not present in current populations and were thus likely lost forever.

Conservation recommendations
Although the genetic patterns in the region are quite complex, lowland populations are genetically different from mountain populations. Lowland populations contain some unique alleles, and alleles which were not found in mountain populations. We found that a lot of genetic diversity is between populations, indicating large population differentiation. Every population of T. calyculata is thus in a way unique and deserves protection and, to save maximum genetic diversity, multiple populations must be protected.
Existence of large viable populations with very low genetic diversity suggests that populations of T. calyculata can be restored even from only a few remaining individuals. Almost no genetic variability was lost at the within-population level during the time period covered by herbarium specimens, i.e., since the 1940s. All this suggests that the recent declines of the populations are not due to loss of genetic diversity.
Mountain populations from the Alps and the Carpathians are not currently endangered. They can, however, be endangered by future climate change due to habitat shifts and change of environmental conditions (Dirnböck et al. 2003). Thus, when setting conservation priorities for this species, mountain populations with high genetic diversity should not be neglected or ignored.