Effect of phosphorus availability on the selection of species with different ploidy levels and genome sizes in a long-term grassland fertilization experiment



  • Polyploidy and increased genome size are hypothesized to increase organismal nutrient demands, namely of phosphorus (P), which is an essential and abundant component of nucleic acids. Therefore, polyploids and plants with larger genomes are expected to be selectively disadvantaged in P-limited environments. However, this hypothesis has yet to be experimentally tested.
  • We measured the somatic DNA content and ploidy level in 74 vascular plant species in a long-term fertilization experiment. The differences between the fertilizer treatments regarding the DNA content and ploidy level of the established species were tested using phylogeny-based statistics.
  • The percentage and biomass of polyploid species clearly increased with soil P in particular fertilizer treatments, and a similar but weaker trend was observed for the DNA content. These increases were associated with the dominance of competitive life strategy (particularly advantageous in the P-treated plots) in polyploids and the enhanced competitive ability of dominant polyploid grasses at high soil P concentrations, indicating their increased P limitation.
  • Our results verify the hypothesized effect of P availability on the selection of polyploids and plants with increased genome sizes, although the relative contribution of increased P demands vs increased competitiveness as causes of the observed pattern requires further evaluation.


Polyploidy, that is, whole-genome duplication, plays an important role in plant speciation and evolution (Grant, 1981; Otto & Whitton, 2000; Ramsey & Schemske, 1998, 2002; Soltis et al., 2003; Rieseberg & Willis, 2007). It is now believed that all flowering plants have experienced at least one round of polyploidy during their evolutionary history (Van de Peer et al., 2009; Jiao et al., 2011), with polyploidy arising, on average, in 35% of species per genus (Wood et al., 2009). Duplicated genes provide a free substrate for mutation and gene evolution, and together with the ability to tolerate extensive genome reorganization, this substrate confers great adaptive and evolutionary potential to polyploids (Soltis & Soltis, 2000; Wendel, 2000; Comai, 2005; Cheng et al., 2007; Otto, 2007; Leitch & Leitch, 2008; Flagel & Wendel, 2009; Parisod, 2012). Along with higher genetic variation, polyploidy is associated with a higher degree of heterozygosity and improved tolerance to selfing (Soltis & Soltis, 2000), which are assumed to explain why polyploids are better colonizers and invaders and why they predominate among widespread and invasive species (Pandit et al., 2011; te Beest et al., 2012).

While the advantages of polyploidy have been widely discussed and acknowledged, certain common disadvantages of whole-genome duplication remain poorly elucidated. The major difficulty for polyploids is probably the initial phase of plant establishment when the newly produced polyploid individual endeavours to establish in a population of dominant diploid ancestor(s). In such a situation, the lone polyploid is effectively suppressed because the limited availability or even absence of mating partners leads to a high proportion of ineffective pollinations and, consequently, the rapid elimination of the polyploid line from the population. This process is known as minority cytotype exclusion (Levin, 1975; Husband, 2000). Polyploidy also results in an immediate increase in the minimum cell size (Cavalier-Smith, 2005), which may either cause physiological or structural changes unfavourable to a particular life strategy or lead to disadvantages within the environment of origin of the polyploid (Stebbins, 1971). For example, trees display polyploidy less frequently than other plants, which may be associated with selection for smaller wood cells and smaller and denser stomata, which are required for the proper mechanical properties of the woody tissue and sufficient stomatal conductance for water and nutrient transport through long xylem pathways (Stebbins, 1938; Beaulieu et al., 2008).

It has also been hypothesized that polyploidy (and increased genome size in general) might be disadvantageous because of increased requirements for nutrients to synthesize the additional chromosomal set(s) and extra DNA (Lewis, 1985; Leitch & Bennett, 2004; Cavalier-Smith, 2005; Leitch & Leitch, 2008). The most dramatic effect on nutrient requirements is expected to involve phosphorus (P) because P is an essential and abundant structural component of nucleic acids, including DNA in chromosomes and rRNA in ribosomes (Sterner & Elser, 2002). P is limiting in many environments (Wassen et al., 2005; Elser et al., 2007; Vitousek et al., 2010), and this limitation is expected to impose several restrictions on the success of polyploids and plants with increased genome size in particular habitats, geographical regions or with specific nutritional strategies (Hessen et al., 2008, 2009; Leitch & Leitch, 2008; Neiman et al., 2009, 2013; Greilhuber & Leitch, 2013). However, to date, the expected effect of P availability on the selection of polyploids, diploids and species with different genome sizes has never been rigorously tested.

In this study, we tested the expected effect of different nutrient treatments and increased soil P availability on the success of polyploids and species with larger genome sizes in the Rengen Grassland Experiment (RGE), which is one of the oldest continuously conducted fertilization experiments worldwide (Schellberg et al., 1999). We measured somatic 2C DNA contents (hereafter referred to as genome sizes) and the ploidy levels of all 74 species present in the RGE and calculated the mean ploidy levels, proportion of polyploids and phylogeny-based means of genome size for vegetation that developed on 30 RGE experimental plots in response to 65 years of continuous application of six fertilizer treatments (untreated control, Ca, Ca + N, Ca + N + P, Ca + N + P + KCl and Ca + N + P + K2SO4). By comparing the effects of the different fertilizer treatments, we aimed to test experimentally whether the long-term supply of nutrients, especially P, may shift the species composition toward polyploids and species with large genome sizes. As changes in the species composition in response to nutrient availability may be closely related to primary CSR species strategies (C, competitors; S, stress tolerators; R, ruderals; Grime, 1977, 2001), these strategies were also considered in the analyses to better understand the observed shifts in the ploidy level and the genome size estimates. Several implications of our findings for the evolution and distribution of polyploids are discussed.

Materials and Methods

The Rengen Grassland Experiment

The RGE was established in 1941 on an oligotrophic pasture grassland in the Eifel Mountains of western Germany at latitude 50°13′N, longitude 6°51′E and an altitude of 475 m (Fig. 1; Schellberg et al., 1999). At the beginning of the experiment, the pasture was cultivated with a grubber and sown with a mixture of productive grasses and legumes. Indigenous species were subsequently allowed to freely recolonize the experimental plots from the surrounding unmanaged vegetation. Since the establishment of the RGE, calcium (burnt lime, CaO), phosphorus (Thomas phosphate, Ca3(PO4)2·(Ca2SiO4)) and potassium (KCl or K2SO4) fertilizers have been applied in single dressings every year, nitrogen fertilizer (ammonium nitrate, NH4NO3) has been applied twice annually, and the area has been mown twice a year. The experimental plots are arranged in five complete randomized blocks, with each block containing five plots 3 × 5 m in size and each plot within a block representing one replication of the five fertilizer treatments (Ca, Ca + N, Ca + N + P, Ca + N + P + KCl and Ca + N + P + K2SO4) (Fig. 1). In 1998, five control plots with no fertilizer application were added to an adjacent area; the added control plots were subjected to the same cutting regime as the original blocks from the start of the RGE but received no fertilizer input (Hejcman et al., 2010a).

Figure 1.

Aerial photograph of the Rengen Grassland Experiment initiated in 1941 and one of the oldest continuous grassland fertilizer experiments with treatments in a properly randomized block design (Schellberg et al., 1999; Hejcman et al., 2010a)

(Photograph: © Michal Hejcman, with permission, June 2005.)

The percentage species cover (abundance) in each plot was recorded visually 65 yr after the beginning of the experiment in late June 2006 (Chytrý et al., 2009). In each 3×5 m plot, the recording was carried out in the central 1.8×3.2 m quadrangle to eliminate edge effects. The analysis of the species abundance data (Chytrý et al., 2009) revealed that three major types of vegetation had developed in response to the fertilizer treatments: in the control plots, submontane short grassland of nutrient-poor soils belonging to the Violion caninae alliance; in the treatments with Ca and without P, montane mesotrophic meadow of the Polygono-Trisetion alliance; and in all the treatments with P application, mesic montane meadow of nutrient-rich soils belonging to the Arrhenatherion alliance. The biomass was sampled for quantitative analyses in early July 2008. Only data from oven-dried plant material are considered here (for details, see Hejcman et al., 2010b). The concentrations of plant-available P (extracted by calcium acetate lactate in the upper 10 cm soil layer) were analysed in all of the plots in 2004 (Hejcman et al., 2010a) (Notes S2). The CSR primary plant strategies (Grime, 1977, 2001) of the species were taken from the BIOLFLOR database (Klotz et al., 2002). The CSR model describes three principal strategies that have developed in plants in response to the combination of two major environmental factors, that is, stress and disturbance (Grime, 1977). The competitive strategy (C) is viewed as adaptation to environments with low stress and low disturbance; the stress-tolerant strategy (S) represents adaptation to high-stress environments with low disturbance; and the ruderal strategy (R) represents adaptation to low-stress but highly disturbed environments; the combination of high stress and high disturbance is assumed to be lethal for plants. For the five taxa for which CSR strategy data were not available in the BIOLFLOR database, we estimated the strategies based on our own long-term field experience with the species: Cerastium holosteoides (CR), Hieracium subg. Hieracium (CSR), Pimpinella saxifraga (CS), Tragopogon pratensis (CSR), and Viola canina (CS). The species taxonomy and nomenclature followed Wisskirchen & Haeupler (1998).

Figure 2.

Mean genomic parameters of species in plots with different nutritional treatments (different symbols) in relation to the concentration of plant-available soil P. The estimates are calculated either by assuming only species presence/absence in the plot (presence/absence data) or by weighting each species presence by its relative biomass content in the plot (dry matter-weighted data). The mean estimates of genome size are phylogeny-corrected. Treatments that do not differ significantly (> 0.05) are marked with the same lowercase letter.

Plant material, genome size and ploidy level measurements

One healthy, mature plant was sampled as a representative of each species in the RGE in May 2011 or June 2012. The absolute genome size (2C DNA content) of the sampled plants was measured using flow cytometry (Doležel et al., 2007) in the Laboratory of Flow Cytometry, Department of Botany and Zoology, Masaryk University. Samples were prepared according to the two-step procedure described by Otto (1990). A piece of fresh leaf was chopped using a sharp razor blade together with an internal standard in a Petri dish containing 1 ml Otto I buffer (0.1 M citric acid, 0.5% Tween 20). The crude nuclear suspension was filtered through a 50 μm nylon mesh. Finally, 1 ml of Otto II buffer (0.4 M Na2HPO4·12 H2O) supplemented with propidium iodide (PI) (50 μg ml−1 final concentration) was added to the filtered suspension. The measurements were performed using a CyFlow ML flow cytometer (Partec, Münster, Germany) equipped with a solid state green laser (532 nm, 100 mW, Cobolt Samba, Solna, Sweden). Fluorescence intensity data were measured for at least 5000 standard + sample nuclei; the coefficients of variance for individual peaks generally ranged between 1.6 and 4.5, with some exceptions in species with high contents of interfering metabolic compounds (e.g. Potentilla erecta, Crepis biennis, Platanthera bifolia). The internal standards and their genome sizes followed Veselý et al. (2012), that is, Carex acutiformis (2C = 800 Mbp), Solanum lycopersicum ‘Stupicke polní tyčkové rané’ (2C = 1697 Mbp), Bellis perennis (2C = 3090 Mbp), Pisum sativum ‘Ctirad’ (2C = 7841 Mbp), and Vicia faba ‘Inovec’ (2C = 23 273 Mbp). Measurements of each plant (species) were repeated on three different days and were then averaged (Table S1).

The ploidy levels of the sampled plants were determined by comparing the measured absolute 2C genome sizes with the data in the Plant C-value database (Bennett & Leitch, 2010) and the available karyological literature (Table S1). The reported ploidy levels refer to the expected genome copy number as a function of the basal chromosome number in a genus. In the agmatoploid genus Carex, which features holocentric (fragmentable) chromosomes, the ploidy levels follow the analysis by Lipnerová et al. (2013), who considered the observation of simultaneous shifts in chromosome number and nuclear DNA content between closely related taxa to indicate the occurrence of polyploidy. In seven species for which different ploidy levels are frequently reported in the literature (Anthoxanthum odoratum, Bromus inermis, Festuca pratensis, Festuca rubra, Galium mollugo, Leucanthemum vulgare, and Trifolium pratense), the homogeneity of the ploidy levels in the RGE was tested through the rapid screening of the ploidy level with 4,6-diamidino-2-phenylindole (DAPI) staining (e.g. Šmarda et al., 2005) in two to five individual plants sampled randomly in different RGE treatments. The procedure followed the method used for PI measurements, except that Otto II buffer with DAPI (4 μg ml−1 final concentration) was used instead of PI, and only external standardization (Bellis perennis) was used. The test confirmed the constancy of the ploidy levels in all the tested species. Accordingly, for all subsequently examined species, only one individual was always selected to be representative of the species for the genome size measurements. Seven species recorded by the vegetation mapping in 2006 were not found in the RGE plots at the time of sampling for flow cytometry analyses in 2011 and 2012. The DNA contents of these species were measured in specimens of the same species sampled outside the RGE, thereby carefully controlling for the taxonomic identity of the measured material and the correspondence between observed ploidy levels and that expected for the species in western Germany (Wisskirchen & Haeupler, 1998).

Species phylogeny

To test the hypothesis of P limitation of polyploids in a phylogenetic framework, we constructed a phylogenetic tree of all 74 species recorded in the RGE. The tree was based on the concatenated alignments of five sequence markers (matK, rbcL, atpF-atpH, rps16 and trnL-trnT) searched for in GenBank (Benson et al., 2012) (Table S1). Owing to a lack of sequence data for Hieracium subg. Hieracium sp. and Platanthera bifolia, these taxa were represented by sequences available for two closely related taxa, Hieracium sabaudum and Platanthera chlorantha, respectively. The highly variable matK and atpF-atpH sequences were aligned using PRANK (Loytynoja & Goldman, 2005, 2008), which is accessible at the Guidance web server (Penn et al., 2010a). The reliability of the resulting alignments was then examined with the Guidance algorithm (Penn et al., 2010b). Only those columns with reliability higher than the default cutoff of 0.93 were used to construct the tree. The other three less variable markers (rbcL, rps16 and trnL-trnT) were aligned with the Clustal algorithm in MEGA 5.0 (Tamura et al., 2011). The gaps were treated as missing data. The tree topology and branch lengths were estimated with the maximum-likelihood method in MEGA 5.0 (Tamura et al., 2011) using the default settings of the parameters and 500 bootstrap replicates. The final topology (Notes S1) agreed well with the hypothesized relationships among the taxa presented by the APG website and the APG III flowering plants phylogeny (Angiosperm Phylogeny Group, 2009).

Statistical analyses

The following parameters were estimated for each plot: the mean absolute 2C DNA content; the mean ploidy level; and the percentage of polyploids. These parameters were estimated for each plot considering the presence or absence of the species (the presence/absence data) and by weighting with the dry mass of the species (the dry mass-weighted data, calculated by multiplying the fraction of the plot covered with a given species by the total dry mass yield of the plot).

Before the parameter estimations, we tested the phylogenetic dependence of DNA content and ploidy level data. This dependence was evaluated by calculating the phylogenetic signals of these variables using the phylosignal function in the picante package (Kembel et al., 2010) in R (R Development Core Team, 2012). These tests demonstrated a significant phylogenetic dependence of DNA content (P < 0.05). This finding indicates that closely related taxa are likely to have similar DNA contents. As response to nutrient treatment may also show a phylogenetic dependence (because of sharing traits other than genome size which may be selectively important in a given treatment, e.g. symbiosis with nitrogen-fixing bacteria in Fabaceae; Franche et al., 2009), the mean per-plot estimates of DNA content must be phylogeny-corrected to allow for evolutionary meaningful (phylogeny robust) comparison among treatments. We used the phylogenetic general least-squares (pgls) procedure (Grafen, 1989) to calculate the phylogeny-independent mean per-plot DNA contents. This calculation was performed using the gls function of the nlme package (restricted maximum-likelihood method; Pinheiro et al., 2012) in R. Assuming that the evolution of DNA content follows the Brownian motion model of evolution, pgls analyses were performed with a Brownian motion-based covariance structure, calculated using the corBrownian function of the ape package (Paradis et al., 2004). Before these analyses, the DNA contents were log10-transformed to produce a normal distribution (Shapiro–Wilk test, P > 0.05), as expected for characters evolving under the Brownian motion model of evolution (Garland et al., 1992). For the pgls analyses, a cover value of 0.000 001 was given to all of the absent species to allow for the consistent calculation of mean per-plot DNA contents in all of the plots with the same phylogenetic tree.

First, we tested the pure effects of the application of a particular nutrient on the estimates of mean genome size, ploidy level and percentage of polyploids (Table 1). This test was performed with linear model (lm) or general linear model analyses (glm; see next paragraph for details) with the nesting of nutrients applications (binary variables; nutrient applied or not) corresponding to the specific additive design of RGE fertilizer treatments (i.e. Ca(N(P(KCl, K2SO4))); Table 1). Because previous analyses did not reveal a substantial effect of the experimental block on the species composition (Chytrý et al., 2009), we examined only the effect of fertilizer application, with all the plots in a given treatment considered as random replicates. In addition to the tests of the effects of individual nutrient applications, lm and glm analyses were used to test for differences in the parameter estimates of plots between treatments, applying user-defined orthogonal contrasts (analogous to application of post hoc multiple comparisons; Fig. 2).

Table 1. Significance (P-value) of the pure effects of the application of a particular nutrient on the observed mean per-plot genomic parameters (Fig. 2)
Nutrient application effectPresence/absence dataDry mass-weighted data
Log10 2CPloidy levelPolyploids (%)Log10 2CPloidy levelPolyploids (%)
  1. Significant results are shown in bold. For detailed results see the Notes S2.

Ca (N(P(KCl, K2SO4)))0.860 0.011 0.5540.8500.122 0.016
N (P(KCl, K2SO4)) 0.002 0.0660.3070.704 0.004 < 0.001
P (KCl, K2SO4) < 0.001 0.001 0.0570.256 < 0.001 < 0.001
KCl 0.2380.5630.8070.6720.6630.921
K 2 SO 4 0.5820.4470.8950.8800.6050.122

After demonstrating the highly significant effect of P application and the significant difference in mean per-plot estimates between P and non-P treatments (Table 1, Fig. 2), several analyses were designed to further determine the reasons underlying this difference. In the subsequent analyses, we divided the species into categories according to their genome size (43 species with small genomes with 2C < 5 Gbp and 31 species with large genomes with 2C > 5 Gbp), ploidy level (diploids and polyploids), and primary CSR strategy (Grime, 1977, 2001) and then compared the species numbers or the total dry matter content of the species in each of these categories between the P-treated and P-untreated plots (Figs 3, 4; Table 2, Notes S2). The calculations for the species with combined primary strategies were performed using a weight proportional to the number of strategies (i.e. a CSR species with 1/3 C, 1/3 S and 1/3 R). As the small and large genome size categories were phylogenetically clustered (test of phylogenetic signal; P < 0.05), the estimates of mean species numbers per plot and total dry matter content for these categories were phylogeny-corrected. This correction was performed as described earlier by conducting mean per-plot estimations of DNA content. The estimates obtained with and without phylogenetic correction were very similar, with no principal effect on the results of the subsequent statistical analyses.

Table 2. Analysis of deviance table with the results (P-values) of general linear model (glm) analyses testing the difference in species numbers and dry matter content in plots between groups of either diploids and polyploids (simple sum per plot) or species with small (< 5 Gbp) and large (> 5 Gbp) genome size (phylogeny-corrected sum per plot) in response to P treatment (P-treated vs P-untreated plots) and the primary plant CSR (C, competitors; S, stress tolerators; R, ruderals) strategy
Response variableGroup=
Small vs large genomesDiploids vs polyploids
Species no.Dry matterSpecies no.Dry matter
  1. Significant results are shown in bold. For detailed results, see the Notes S2.

P treatment < 0.001 ≪ 0.001 < 0.001 ≪ 0.001
Group ≪ 0.001 ≪ 0.001 < 0.001 0.269
CSR ≪ 0.001 ≪ 0.001 ≪ 0.001 ≪ 0.001
P treatment × group0.216 < 0.001 0.224 ≪ 0.001
P treatment × CSR ≪ 0.001 ≪ 0.001 < 0.001 ≪ 0.001
Group × CSR ≪ 0.001 ≪ 0.001 0.061 < 0.001
P treatment × group × CSR 0.012 ≪ 0.001 0.625 ≪ 0.001
Figure 3.

Species number and sum of above-ground dry mass in experimental plots for diploids and polyploids and species with smaller and larger genome sizes in relation to plant-available soil P content.

Figure 4.

Difference between P-unfertilized (nonP) and P-fertilized plots (P) in species number and sum of above-ground dry mass in experimental plots for diploids and polyploids and species with smaller and larger genome sizes considering the species primary CSR (C, competitors; S, stress tolerators; R, ruderals) strategy. Boxplots show the median (thick horizontal line), interquartile range (boxes), nonoutlier range (whiskers) and outliers (circles).

All tests of treatment effects and of differences in parameters between treatments or species were conducted in the R program with linear models (function lm) for normally distributed response variables (mean per-plot log10 2C estimates and mean per-plot ploidy level estimates). In the case of over- or underdispersed or heteroscedastic data, general linear models were applied with the appropriate distribution model and link functions: percentage of polyploids, quasibinomial distribution with logit link; dry matter content, gamma distribution with inverse link (both with the glm function); and species numbers, negative binomial distribution (function glm.nb of the MASS library). When the effect of the plant CSR strategy on a tested parameter was statistically significant, the differences among the individual strategies were compared with planned contrasts. The correlation between DNA content and ploidy level was tested with pgls and the Brownian motion-based covariance structure using the R packages and functions described earlier.


The genome size of all 74 species in the RGE varied from 2C = 662 Mbp in Luzula campestris to 2C = 42 620 Mbp in Platanthera bifolia. Of the 74 measured species, 41 were diploid and 33 were polyploid (two triploid, 24 tetraploid, four hexaploid and three octoploid). The application of P (three treatments) displayed the most dramatic effect on the observed mean per-plot estimates (five estimates for each of the six treatments, i.e. 30) in the statistical analyses (Table 1, Fig. 2, Notes S2). This result is most apparent in the comparison of mean ploidy level of species in the plot, which was significantly higher in the P-treated than in the P-untreated plots. The increased P limitation of polyploids was further demonstrated by the decrease in mean ploidy and polyploid percentage in non-P treatments, where the P deficiency was further increased with application of N and Ca (Table 1, Fig. 2, Notes S2); N is known to enhance the deficiency of plant-available P in the soil as a result of more rapid P cycling and increased P removal by the harvested biomass, and Ca enhances the deficiency by binding free P to form an insoluble and plant-unavailable complex (Gress et al., 2007; Hejcman et al., 2010a; Vitousek et al., 2010). Similarly to the effects on polyploids, application of N led to a decrease while application of P led to an increase in the mean per-plot genome size. However, this difference was statistically significant only when calculated with the species presence/absence data (Fig. 2, Table 1).

To analyse the reasons for the observed differences, we further compared the effect of P treatment on the species number, dry mass weight, and frequency of the CSR strategies between the diploids and polyploids, and between the species with small and large genome sizes (Table 2, Figs 3, 4, Notes S2). In general, P-treated plots were less species-rich and produced more above-ground dry matter biomass (Fig. 3). The decrease in the species number in the P treatments was primarily caused by a decline in the stress tolerators and the ruderals, and the increase in the herbage yield was caused by an increased above-ground biomass of the competitors (Fig. 4).

The polyploids were less species-rich than the diploids in all the CSR strategies, but the sum of the herbage yields was higher for the polyploids than for the diploids (Fig. 4, Table 2). The analysis revealed that the increased mean ploidy level and proportion of polyploids comprising the biomass in the P-treated plots were mainly caused by a more intensive decline in diploid than in tetraploid competitors (P < 0.001) (Fig. 4b), while the decline of the ruderals and the stress tolerators was similar for both ploidy levels in the P treatments (P = 0.213) (Fig. 4b); and by a greater increase in the total dry matter herbage yield of the polyploids compared with the diploids, which was observable in all CSR strategy categories regardless of their decreased species number (P < 0.001) (Table 2, Fig. 4d). This increase was mainly caused by the establishment of dominance in the P-treated plots by productive and competitive polyploid grasses, such as Trisetum flavescens, Arrhenatherum elatius, Alopecurus pratensis, Bromus inermis, and Dactylis glomerata. These grasses were absent or present only at low abundances in plots with no P application, but their abundance clearly increased in P-rich plots, which have reasoned in the demise of less competitive diploid grasses (Briza media, Helictotrichon pubescens, Nardus stricta) or other species with noncompetitive or combined primary strategies, mainly represented by diploids (Betonica officinalis, Carex pilulifera, C. panicea, Hypochaeris radicata, Lathyrus linifolius, Linum catharticum, Luzula campestris, Ranunculus nemorosus, Rhinanthus minor and Selinum carvifolia; see also Table S1). This result indicates that polyploid competitors responded more strongly to P application and that these polyploids may thus be assumed to be more P-limited than the diploids.

The analyses with genome size were in many aspects similar to the results with the polyploids (Fig. 4, Table 2), because genome size and polyploidy were closely related in the RGE species dataset (pgls, P = 0.011). Similar to the polyploids, plants with a larger genome size were mostly competitors (P ≪ 0.001), with few ruderals or stress tolerators (Fig. 4a, Table 2). In contrast to the polyploids, the results with the genome sizes differed only when comparing the total biomass production, which sharply declined in the large genome species with a ruderal or stress-tolerating strategy in the P-treated plots (Fig. 4c), while in the polyploids, the total biomass production increased in the P-treated plots for all CSR strategies (Fig. 4d). This pattern was primarily the result of the complete disappearance of several large-genome diploid species in the P-treated plots. These species combined competitive and stress-tolerant strategies and were dominant species in the plots without P application: Briza media (CSR), Danthonia decumbens (CS), Lathyrus linifolius (CSR), Ranunculus nemorosus (CS), and Selinim carvifolia (CS) (Table S1).


The results of this study clearly indicate that the long-term application of P to oligotrophic grassland substantially increased the relative success of polyploids and, indirectly, species with larger genome sizes. This study provides the first experimental support for the expected effect of plant-available soil P on the frequency of polyploids in plant communities (Leitch & Leitch, 2008; Greilhuber & Leitch, 2013). However, it seems clear that the increased P limitation of polyploids is not only caused by the expected increased demand for extra DNA processing but is also related to the increased competitiveness of polyploids. This competitiveness may be associated with the effect of extra DNA on the increase in cell size or the process of polyploid establishment. In spite of the uncertainty regarding the relative importance of particular demands and processes in causing increased P limitation of polyploids, this limitation may be viewed as an important mechanism that might effectively counterbalance the genetic advantages of polyploids and prevent their unlimited spread and ubiquitous dominance.

The success of polyploids in P-rich plots – a consequence of higher overall competitiveness

Our analysis has shown that the response of species to P application is strongly determined by their primary CSR strategy, with competitive species prevailing in the P-rich plots (Table 2, Fig. 4, Notes S2). This finding is similar to the results of other long-term fertilizer experiments established in soils with low P availability, in which the addition of P mainly increased the success of competitors at the expense of stress-tolerant and ruderal species (Galka et al., 2005; Silvertown et al., 2006; Hejcman et al., 2007; Semelová et al., 2008). This result suggests that the effects of P enrichment or P deficiency on the success of polyploids in other communities would greatly depend on the proportion of polyploids in the local pool of competitor species or, more generally, in a pool of species with other locally advantageous life strategies. Hypothetically, the P enrichment of a site in which the majority of the local species pool consists of diploid competitors might simply result in a relative decrease in polyploids. However, several results indicate that such situations are probably rare, because competitiveness and polyploidy seem to be closely related.

First, the competitive strategy clearly predominated among the polyploid species in the RGE (Fig. 4b, Table 2), and this pattern may occur among the herbaceous flora of larger regions, such as the flora of Sheffield, UK (Hodgson, 1987). Secondly, polyploids became dominant and produced more biomass in the RGE when they were supplied with sufficient P, although there were lower absolute numbers of polyploids than diploids in the pool of competitor species (Fig. 4b,d). The higher individual dry mass yield of the polyploid species and their ability to establish dominance under favourable P conditions indicate that polyploids must be stronger competitors than their diploid relatives at P-rich sites. As discussed in the following section, the increased competitiveness of polyploids may originate from a close relationship between genome size and cell volume or from the selection of newly arising polyploids during their establishment in populations of their diploid ancestors.

Extra DNA – a P-demanding preadaptation for stronger competitiveness?

The fact that the success of polyploid grasses in P-rich plots is associated with their competitive strategy complicates the clarification of the extent to which the observed relative P limitation of polyploids is the result of the increased demands for extra DNA processing or a general feature associated with the competitive life strategy. However, the effects of extra DNA and competitiveness on the increased P limitation in polyploids would be difficult to separate, as they may be closely or mutually conditioned.

Across organisms, the size of the nucleus (DNA content) is closely correlated with the amount of cytoplasm and, correspondingly, the cell volume (Cavalier-Smith, 2005). Because DNA comprises only a limited fraction of the total cellular dry mass and cellular P content (Lynch, 2007), the production of larger cells may be dramatically more P-demanding than DNA replication. Thus the effect of extra DNA on P demand may not result from the requirement of additional P to build in extra DNA but rather from the requirement of P to build additional cellular components and ribosomes (RNA). P availability is then limiting for the speed of cell division and growth (Elser et al., 2003; Hessen et al., 2009), which are known to determine species success, particularly in productive environments (Grime, 2001).

Larger cells need not have a purely negative effect on increased nutrient demands; they may also allow for larger body size. Indeed, the increased size of polyploids compared with their diploid relatives is frequently observed in the taxonomic literature (Stebbins, 1971). The ability of larger plants to shade their neighbours may provide polyploids with important advantages in competition for light, one of the major processes determining species success in a productive community (Hautier et al., 2009). In this respect, genome doubling may be viewed as a preadaptation through which polyploids become successful light competitors, but this advantage would only be realized in the presence of a nutrient load sufficient for rapid growth. In the RGE, this may be the case for tall, fast-growing, polyploid grasses such as Arrhenatherum elatius (Grime, 2001), which is nearly absent in the non-P-treated plots but dominates in all P-treated plots.

If increased competitiveness were solely an outcome of cell size, then we should observe increased competitive performance in all plants with large genomes. Indeed, our data show that P limitation may be associated with larger genome size. However, this association is largely driven by the close correlation of genome size and ploidy level. Of diploids with larger genomes (2C > 10 Gbp; Briza media, Lathyrus linifolius, L. pratensis, Listera ovata, Platanthera bifolia, and Ranunculus nemorosus), only L. pratensis occurred sporadically in plots with high P availability. Of these six diploids, Lpratensis is the only species with a purely competitive strategy, while the others are classified as partial stress tolerators. The failure of diploids with large genomes to become established in P plots indicates that extra DNA and larger cell size do not inevitably play a dominant role in determining species competitiveness. If this finding holds true, it would also imply that the improved competitiveness of polyploids at P-rich sites must have evolved through additional mechanisms beyond increases in cell and body size.

Increased polyploid competitiveness – an outcome of initial selection from diploid ancestor populations?

Several models describe theoretical conditions under which newly formed polyploids can establish in competition with one or both of their parental species (Soltis et al., 2003). It is hypothesized that polyploids may persist only when they outcompete parental diploids or when diploid parents are stochastically lost because of their small population size (Fowler & Levin, 1984). Furthermore, it is suggested that polyploids may also coexist with their parents as a result of immediate habitat differentiation (Fowler & Levin, 1984; Rodriguez, 1996) or, in more general terms, by the development of effective interploidy reproductive barriers, thus avoiding the effect of minority cytotype exclusion (Husband et al., 2013). The initial deficiency (absence) of mating partners may also be overcome by an increased tolerance to selfing and asexual reproduction and by prolonging the time available to search for partners (Stebbins, 1971). Consequently, polyploids are rare among outcrossing annuals and occur most frequently among polycarpic selfing perennials (Gustafsson, 1948; Stebbins, 1971; Otto & Whitton, 2000; Barringer, 2007). As an analogy to the need for effective mating, the need for a competitive advantage over diploid parents might function as a selective filter that allows for the establishment of only polyploids with high competitive performance. This process would naturally lead to an increased proportion of highly competent competitors among established polyploid species.

Assuming that diploid and polyploid species would initially have very similar ecological and resource-usage strategies, the selected competitive trait might be the speed of resource utilization and growth which can be generally selected for by intraspecific competition (Begon et al., 2006). The hypothesis that polyploids exhibit higher competitive performance is, however, still indirectly supported primarily by the high invasion capacity of numerous neopolyploids, which is expected to be greatly dependent on competitive ability (te Beest et al., 2012). Experimental demonstrations of the competitive dominance of natural neopolyploids over their diploid ancestors are still rare (Maceira et al., 1993; Collins et al., 2011).

Could P limitation affect the worldwide distribution of polyploids and polyploidy?

In addition to affecting the polyploid frequency in particular plant communities, the P limitation of polyploids might be expected to influence the distribution of polyploids across larger geographical scales (Leitch & Leitch, 2008; Neiman et al., 2013; Greilhuber & Leitch, 2013). Namely, polyploids may be expected to be relatively infrequent in P-limited environments, which handicap polyploids as a result of their high P demands for DNA and cell synthesis. However, polyploids may be expected to excel in P-saturated sites, where they may profit from their robust growth and the advantageous competitive traits selected during their establishment in ancestral diploid populations. However, predictions concerning the relation between soil P and the abundance of polyploid taxa at larger geographical scales and in different vegetation types would be more complex and involve additional factors that determine polyploid success in the RGE and grasslands in general.

On a wider geographical scale, two particular examples fit well with the expectations described. It has long been known that the frequency of polyploids is relatively low in certain tropical regions, such as West Africa (Morton, 1966; Baquar, 1976; Grant, 1981), or in the fynbos biome of South Africa (Gold-blatt, 1978). Both these ecosystems are strongly P-limited because of old, P-leached soils (Porder et al., 2007; Lambers et al., 2008; Vitousek et al., 2010) or the abundance of soils derived from low-P bedrock (Witkowski & Mitchell, 1987), respectively. However, there are also alternative explanations; in particular, the infrequent polyploidy in both regions is generally associated with the predominance of woody species, in which polyploidy might be limited by structural and physiological constraints on cell size (Stebbins, 1938; Beaulieu et al., 2008). Additionally, the high proportion of allopolyploids in P-poor environments in arctic regions indicates that the percentage of polyploids in a given area is not necessarily linked to the availability of soil P. In these environments, polyploidy seems advantageous because of fixed heterozygosity and improved tolerance to selfing, which assists polyploids in colonizing and surviving harsh climates where mating partners are rare (Brochmann et al., 2004).


The authors are grateful to the four reviewers for their useful comments in the preparation of the final draft of this paper. This research was supported by the Czech Science Foundation (projects GACR P505/11/0881 and GACR 505/12/1390) and the German Research Foundation (DFG) (project SCHE 549/2-1).