Morphological and environmental differentiation as prezygotic reproductive barriers between parapatric and allopatric Campanula rotundifolia agg. cytotypes

Abstract

Background and Aims

Reproductive isolation and local establishment are necessary for plant speciation. Polyploidy, the possession of more than two complete chromosome sets, creates a strong postzygotic reproductive barrier between diploid and tetraploid cytotypes. However, this barrier weakens between polyploids (e.g. tetraploids and hexaploids). Reproductive isolation may be enhanced by cytotype morphological and environmental differentiation. Moreover, morphological adaptations to local conditions contribute to plant establishment. However, the relative contributions of ploidy level and the environment to morphology have generally been neglected. Thus, the extent of morphological variation driven by ploidy level and the environment was modelled for diploid, tetraploid and hexaploid cytotypes of Campanula rotundifolia agg. Cytotype distribution was updated, and morphological and environmental differentiation was tested in the presence and absence of natural contact zones.

Methods

Cytotype distribution was assessed from 231 localities in Central Europe, including 48 localities with known chromosome counts, using flow cytometry. Differentiation in environmental niche and morphology was tested for cytotype pairs using discriminant analyses. A structural equation model was used to explore the synergies between cytotype, environment and morphology.

Key Results

Tremendous discrepancies were revealed between the reported and detected cytotype distribution. Neither mixed-ploidy populations nor interploidy hybrids were detected in the contact zones. Diploids had the broadest environmental niche, while hexaploids had the smallest and specialized niche. Hexaploids and spatially isolated cytotype pairs differed morphologically, including allopatric tetraploids. While leaf and shoot morphology were influenced by environmental conditions and polyploidy, flower morphology depended exclusively on the cytotype.

Conclusions

Reproductive isolation mechanisms vary between cytotypes. While diploids and polyploids are isolated postzygotically, the environmental niche shift is essential between higher polyploids. The impact of polyploidy and the environment on plant morphology implies the adaptive potential of polyploids, while the exclusive relationship between flower morphology and cytotype highlights the role of polyploidy in reproductive isolation.

INTRODUCTION

Polyploidy, the possession of more than two complete chromosome sets gained through whole-genome duplication (WGD), is present in all living angiosperms (Jiao et al., 2011) and is widely considered one of the most important and common modes of plant speciation (Weiss-Schneeweiss et al., 2013; Wendel, 2015; Landis et al., 2018). Understanding the indisputable evolutionary role of polyploidy and the ubiquity of polyploids in nature (Soltis and Burleigh, 2009; Mandáková et al., 2017) is challenging because newly formed polyploids as a minority cytotype face frequency-dependent selection that can lead to their extinction (Levin, 1975). The complex physiological processes needed for the successful survival and establishment of polyploids and the necessity of reproductive isolation from their diploid progenitors thus raised the hypothesis of polyploids as dead ends (Soltis et al., 2009; Mayrose et al., 2015; Levin, 2019). Reproductive barriers are crucial to reduce the risk of insufficient fertilization (López-Jurado et al., 2019) and/or the production of unfit hybrids (Hopkins, 2013). Polyploidy itself instantly creates an effective postzygotic reproductive barrier between diploids and related polyploids (tetraploids) by producing undeveloped, less viable or sterile triploid offspring (triploid block; Köhler et al., 2010; Hülber et al., 2015). However, the triploid block is probably not sufficient to prevent ineffective hybridization, resulting in partially fertile triploids (triploid bridge; Mandáková et al., 2013; Suda and Herben, 2013; Lafon-Placette et al., 2017; Mao et al., 2020) or tetraploids formed by fusions of reduced and unreduced gametes (Sutherland and Galloway, 2016). Therefore, prezygotic reproductive barriers are important because they promote assortative mating within each cytotype (Kolář et al., 2017; Castro et al., 2020).

Altering genome dosage, WGD immediately impacts cell size (nucleotypic effect; Levin, 1983) and plant phenotype (Gigas effect; Ramsey and Schemske, 1998; Simón-Porcar et al., 2017). The morphology often changes predictably, especially in synthetic neopolyploids, which may, however, differ morphologically from polyploids already established in natural populations under selection (Husband et al., 2016). Altered floral reproductive traits often do not have an impact on pollinator behaviour or plant self-incompatibility (Porturas et al., 2019; Castro et al., 2020). Thus, cytotype spatial segregation often accompanied by environmental niche differentiation explains most patterns of reproductive isolation (López-Jurado et al., 2019; Castro et al., 2020).

Polyploid niche evolution is limited by available niche space (Brochmann et al., 2004) and may act not merely at the ploidy level but also at different lineages within the same cytotype (López-Jurado et al., 2019). Despite a possible lack of niche differentiation between diploids and polyploids due to phylogenetic niche conservatism (Glennon et al., 2014) or their recent origin and ongoing gene flow (Wos et al., 2019), niche expansion is frequent for triploids and tetraploids. As the first polyploids formed, they fill the niche space unoccupied by diploids, thus avoiding competition and interploidy mating. For subsequent polyploids, however, environmental divergence may be necessary to find available niches, and the remaining niche space is often at the extremes of environmental gradients. Higher ploidies may then encounter constraints to the range due to limitations in their environmental tolerances (Laport et al., 2013; Muñoz-Pajares et al., 2017; López-Jurado et al., 2019). When compared to their diploid or lower polyploid ancestors, polyploids show niche contraction because they are locally adapted to narrower and marginal niches in specific habitats (Sonnleitner et al., 2016; López-Jurado et al., 2019).

Extreme climatic conditions and environmental stress may also enhance the formation of unreduced gametes (Ramsey and Schemske, 1998; Wilson et al., 2020), leading to recurrent polyploid formation. Local adaptations of multiple lineages within the same cytotype (Parisod and Besnard, 2007; Castro et al., 2018) often lead to a wider polyploid distribution (López-Jurado et al., 2019; Castro et al., 2020; Wilson et al., 2020). Subfunctionalization and neofunctionalization of the duplicated genes (Moore and Purugganan, 2005; Flagel and Wendel, 2009) may be essential for local adaptation and allow for colonization of new habitats, which may lead to invasiveness (te Beest et al., 2012).

Successful polyploid establishment often requires more than just size-related changes in plant phenotype (Porturas et al., 2019). Phenotypic changes driven by the local environment promote higher competitive abilities or habitat displacement (Balao et al., 2011; Laport et al., 2017). However, they also mask the morphological changes induced by polyploidization. Distinguishing the effect of polyploidization from the effects of selection and local adaptation that follow the reproductive isolation of polyploids from their diploid progenitors is challenging and crucial for studies on plants from natural populations.

Heteroploid taxa provide useful insight into the evolution of different ploidy levels (Kolář et al., 2017) and allow the assessment of whether differentiation in environmental requirements and morphology proportionally increases with increasing ploidy level, following the hypotheses of niche expansion and the Gigas effect, or is individual for each cytotype. Since ploidy level and cytotype spatial isolation may influence the strength of the prezygotic reproductive barriers, we compared cytotype pairs based on their ploidy level and geographical distribution. We hypothesize that cytotypes in contact (sympatry, parapatry) have higher morphological and ecological differentiation driven by selection for assortative mating (reinforcement; Hopkins, 2013). On the other hand, morphological and environmental differentiation between populations with limited contact (parapatry or allopatry) is hypothesized to be driven by local ecological conditions in different areas.

Three cytotypes (2x, 4x, 6x) from a polyploid complex of Campanula rotundifolia agg. co-occur at different levels of contact in Central Europe (the Bohemian Massif, the Pannonian Basin and the Western Carpathians; Mráz, 2005). A contact zone is either present within a population (sympatry, mixed-ploidy populations of 2x+4x and 4x+6x; Kovanda, 1967, 1970a, 2002) or between uniform-ploidy populations of each cytotype (parapatry; Kovanda, 1966; Mráz, 2005; Rauchová, 2007; Šemberová, 2013). The known cytotype distribution pattern shows a longitudinal shift from diploids in the west (the Bohemian Massif) to tetraploids in the east (2x–4x mixed ploidy populations in the Bohemian Massif and a continuous range of tetraploids in the Pannonian Basin and the Western Carpathians) (see first figure in Supplementary Data Material S1). Hexaploids are reported mainly from 4x–6x mixed ploidy and a few uniform ploidy populations almost exclusively in the Pannonian Basin (Gadella, 1964; Kovanda, 1967, 1970a, b, 1983, 2002; Mráz, 2005).

We aimed: (1) to revise and complement existing data on the distribution of 2x, 4x and 6x cytotypes; (2) to detect differences and variation in environmental requirements and morphological differentiation between parapatric and allopatric cytotype pairs and to test whether the changes proportionally increase with each ploidy level, following the hypotheses of niche expansion and the Gigas effect, or are individual for each ploidy level; (3) to quantify the extent of morphological variability accounted for by the local environmental conditions and cytotype effects; and (4) to infer the contribution of environmental and morphological differentiation to prezygotic reproductive barriers between the three cytotypes.

MATERIALS AND METHODS

Study group and study area

The three cytotypes (2n = 2x = 34 chromosomes, 2n = 4x = 68 chromosomes, and 2n = 6x = 102 chromosomes) of Campanula rotundifolia agg. (harebells, Campanulaceae, also referred to as Campanula sect. Heterophylla sensuMansion et al., 2012) are widely distributed in Central Europe, an area where mixed-ploidy, as well as uniform-ploidy, populations are described (Supplementary Data Material S1; Gadella, 1964; Kovanda, 1967, 1970a, 1977, 1983, 2002; Mráz, 2005; Rauchová, 2007). While diploids were identified in the Bohemian Massif (western part of the Czech Republic), tetraploids inhabit the broadest geographical range consisting of two disjunct areas, one in the Bohemian Massif and the other in the Western Carpathians (north-eastern part of the Czech Republic, central and eastern Slovakia), with overlap into the northernmost part of the Pannonian Basin (south-eastern part of the Czech Republic, north-eastern Austria, southern Slovakia). Hexaploids are found in the north-western part of the Pannonian Basin and the adjacent part of the Western Carpathians.

Central European harebells are usually considered autopolyploids (Kovanda, 1970b; Laane et al., 1983; Sutherland and Galloway, 2018) with complicated interspecific relationships, unresolved phylogeny (Mansion et al., 2012; but see also Nierbauer et al., 2017; Sutherland et al., 2018; Wilson et al., 2020) and unclear taxon delimitation (Kovačić, 2004), often based on ploidy-related changes in size-affected phenotypic traits (stomata, pollen). However, the trend of size increase is not beyond doubt (Kovanda, 2002; Rauchová, 2007), and inferring polyploid origins from morphology only or combined with niche or distributional range has pitfalls in merging evolutionarily independent lineages (Segraves et al., 1999; Doyle and Sherman-Broyles, 2017). Thus, similar to works in other polyploid complexes (Sonnleitner et al., 2016; Hanušová et al., 2019; López-Jurado et al., 2019), we studied cytotype environmental and morphological differentiation while omitting taxonomic entities (diploids and tetraploids in the Bohemian Massif are represented by C. rotundifolia L. and C. gentilis Kovanda, tetraploids from the Western Carpathian and hexaploids by C. moravica(Spitzner) Kovanda; Supplementary Data Fig. S1). In the study region, diploid and tetraploid cytotypes are also present in endemic taxa. These taxa were omitted from the study because of their restricted distribution and strict ecological requirements [subalpine tetraploids: C. bohemica Hruby: Krkonoše Mts, C. gelida Kovanda: Jeseníky Mts, C. rotundifolia subsp. sudetica(Hruby) Sóo: Krkonoše Mts and Jeseníky Mts, C. tatrae Borbás: Tatry Mts, and diploids: C. serrata(Kit. ex. Schult.) Hendrych: Carpathian Mts, C. cochleariifolia Lam.: alpine zone in most European mountains and diploid C. xylocarpa Kovanda from Slovakian karst].

Sampling

Plant material from 231 localities was sampled between 2012 and 2017 (Fig. 1; Supplementary Data Table S1), with a particular focus on localities with published chromosome counts and mixed ploidy localities (Supplementary Data Material S1; Gadella, 1964; Kovanda, 1967, 1970a, 1977, 1983, 2002; Mráz, 2005; Rauchová, 2007, Šemberová, 2013). At each locality, a GPS coordinate was saved, and plants for further analyses were randomly sampled to cover the entire morphological and habitat spectra (e.g. plant height, number of flowers, shaded vs. light stands) at a minimum of 15-m spans to avoid clones (Stevens et al., 2012). On average, 14 ± 18 individuals and 34 ± 19 individuals were sampled on localities without and with previously published chromosome counts, respectively (Table S1 and Supplemetary Material S1). Selected mixed-ploidy localities were subsequently studied to obtain higher resolution data regarding cytotype spatial distribution.

Fig. 1.

Cytotype distribution of Campanula rotundifolia agg. in the investigated area of Central Europe inferred from flow cytometry analyses of individuals from 231 localities in the Czech Republic, Slovakia and Austria. Black dots mark the populations used for morphometric analysis. Geographical features were adapted from Mráz (2005) and simplified.

DNA ploidy (relative genome size) estimation

All individuals were analyzed by flow cytometry (FCM) to estimate the ploidy level following the best practice recommendation (Sliwinska et al., 2021). Sample preparation followed a simplified two-step procedure (Doležel et al., 2007): an appropriate amount of leaf tissue for both Campanula and internal standard (Bellis perennis, 2C = 3.38 pg, Schönswetter et al., 2007b) was chopped together by a razor blade in 0.5 mL of ice-cold Otto I buffer (0.1 m citric acid, 0.5 % Tween 20) in a plastic Petri dish. The suspension of nuclei was filtered through a 42-µm nylon mesh. For ploidy level estimation, 1 mL of staining solution, containing Otto II buffer (0.4 m Na₂HPO₄·12H₂O), 4 µg·mL⁻¹ 4′,6-diamidino-2-phenylindole (DAPI) and 25 µL·mL⁻¹ β-mercaptoethanol, was added to the suspension of nuclei. The solution of stained nuclei was analysed using a Partec CyFlow ML cytometer (Partec GmbH, Münster, Germany) equipped with a 365-nm UV-LED as a source of UV light for DAPI excitation. The fluorescence intensity of at least 3000 particles was recorded for further data processing. Up to 10 individuals were pooled, and samples with a CV (coefficient of variation) >3 % were re-analysed. Altogether, 6011 individuals were screened for DNA ploidy level.

Environmental data

The environmental niches of the cytotypes were defined using 36 variables (Supplementary Data Table S2). The data were acquired from climatic and topographic GIS layers preprocessed by GeoModel Solar (Bratislava, Slovakia), developer and operator of the SolarGIS service. Air temperatures at 2 m were derived from the Climate Forecast System Reanalysis (National Centres for Environmental Prediction, USA) for the period from 1990 to 2009. The data were spatially enhanced to 30 arc-sec resolution by disaggregation based on the correlation between terrain altitude and temperature. Precipitation data were processed from the database of the Global Precipitation Climatology Centre project (Schneider et al., 2014) for the period from 1951 to 2000. Source data resolution was increased to 2 arc-minutes by disaggregation based on the focal correlation of precipitation with SRTM 30 elevation data and cloudiness (clear-sky index) derived from the SolarGIS database. Topographic data (altitude, slope and aspect) were obtained from the terrain elevation model (The Shuttle Radar Topography Mission data – SRTM3) at 15 arc-sec resolution. Photosynthetically active radiation (PAR) was calculated by the SolarGIS model for the period 1994–2013 at 15 arc-sec resolution.

For environmental modelling, long-term yearly and monthly averages of air temperatures and precipitation were used. As a measure of climatic variability, the average annual standard deviations of temperatures and precipitation were calculated. The aspect values were linearized and rescaled to range from 0 to 4 (0 = south, 1 = south-east and south-west, 2 = east and west, 3 = north-east and north-west, 4 = north). Finally, information on geological bedrock was added from online maps (https://mapy.geology.cz/geocr50/, http://apl.geology.sk/gm50js/), the intensity of human influence was estimated using personal observations in the field (human-influenced vs. natural), and habitat type was ranked from rocks (0) through grasslands (1) to forests (2).

Plant morphological data

A subset of 1157 individuals was selected from the sampled material based on the presence of 19 primary characters (Supplementary Data Material S2) chosen according to previous studies (Kovanda, 1970b; Rauchová, 2007). Morphological characteristics were measured on material sampled in July–August 2012 and 2013 using a digital calliper, and the outer part of the ovary was checked for the presence or absence of the papilla by using an Olympus SZ51 stereomicroscope with a magnification of 8–10× (Olympus Corp., Tokyo, Japan).

Data analysis

Environmental niche shifts and morphological differentiation among the cytotypes were assessed using discriminant analysis. However, a high number of environmental variables (36) and morphological variables (29) and strong correlations among them (environmental data: Pearson’s r = −0.937–0.998, morphological data: r = −0.745–0.885) pose severe limitations on the use of traditional methods. Under these circumstances, classical linear discriminant analysis can result in a solution with unstable coefficients or even in the inability to determine the optimal discriminant function (Wehrens, 2011). An ideal way of solving this problem would be to consider only a subset of uncorrelated proximal variables (i.e. those directly influencing the distribution of the cytotypes or describing morphological differentiation). Unfortunately, we had little prior knowledge of the direct effects of habitat characteristics on the cytotype distribution or the biologically meaningful morphological differences between the cytotypes. Moreover, there is no guarantee that individual proximal variables would bear independent (uncorrelated) information (Jiménez-Valverde et al., 2009). Another possibility would be to exclude correlated environmental and morphological variables prior to the analysis. De Marco and Nóbrega (2018) advise against subjective dropping of variables and advocate the use of latent variable methods such as principal component analysis (PCA). The latent variable approach is easily implemented and commonly used in environmental niche modelling (e.g. Broennimann et al., 2012; De Marco and Nóbrega, 2018, and references therein) and multivariate morphometrics (e.g. Blackith and Reyment, 1971; Claude, 2008). We adopted a partial least square (PLS), supervised latent variable method that has an advantage over traditional PCA since it takes the dependent variable into account when defining ordination scores and loadings, whereas PCA captures variance in predictors only. The general idea of PLS is to construct a few latent variables while maximizing the covariance between a predictor and response matrices. PLS in discrimination analysis provides a dimension reduction technique that finds the optimal group separation while being guided explicitly by between-group variability (Barker and Rayens, 2003). However, structured noise may complicate the interpretability of PLS models, particularly when there are many components (Eriksson et al., 2006). To remove such undesirable systematic variation in the data, we employed PLS with orthogonal projection to latent structures (OPLS) (Trygg and Wold, 2002). The objective of OPLS in discriminant analysis (OPLS-DA) is to divide the systematic variation in the predictor matrix into two parts: the predictive part, which models the relationships between variables and groups, and the orthogonal part, which captures the systematic variation in variables that is linearly unrelated to group separation, which generally leads to improved model interpretability.

The OPLS-DA models of environmental and morphological differences among the cytotypes were built with two predictive components to efficiently separate the three ploidy levels. The environmental and morphological data were standardized to z-scores to equalize the weight of variables in the analysis. Since the PLS methods are prone to overfitting, we used cross-validation to select the optimal number of components and to assess the model performance on independent data (Westerhuis et al., 2008). We iteratively fitted OPLS-DA models with increasing complexity (up to 10 components), trained them on all the data except for one observation at a time and made predictions for those data points left out of the training sets (leave-one-out cross-validation). In both environmental and morphological datasets, models with more than two components provided little or no additional predictive power (Supplementary Data Fig. S2). Moreover, the relative importance of the components was assessed using randomization tests where variance explained by the components was compared with its null distribution generated from data randomly reshuffled 1000 times (Manly, 2006; Westerhuis et al., 2008). To facilitate interpretation of the results, OPLS-DA score plots with 95 % confidence ellipses were displayed. Variable weights for the significant components were plotted to assess the relative influence of environmental and morphological characteristics on the discrimination of ploidy levels. Using an orthogonal projection to latent structures, these weights are primarily related to differences between cytotypes and their interpretation is quite straightforward (Wold et al., 2001).

Environmental niche breadth and morphological variation were compared among the three cytotypes using distance-based tests of homogeneity of multivariate dispersions (Anderson, 2006). Environmental heterogeneity was quantified by calculating pairwise Euclidean distances among sampling sites from the matrix of standardized habitat characteristics. For morphological analyses, characters measured across multiple individuals were averaged within sites to cope with autocorrelation in the repeated measurements. The spread of sites from their medians in principal component space was used as a measure of heterogeneity. Equality of multivariate dispersion was tested using a randomization test based on 1000 permutations of the least absolute deviation residuals.

In addition to the overall comparison, cytotypes were divided into subsets based on their geographical distribution, and the differences in environmental niche and morphological traits were compared between parapatric and allopatric populations. In parapatry, we compared 2x–4x from the Bohemian Massif, 6x–4x from the Western Carpathians and 2x–6x from the Bohemian Massif and the Pannonian Basin. In allopatry, we compared 2x–4x from the Western Carpathians, 6x–4x from the Bohemian Massif and 4x–4x from the Bohemian Massif and the Western Carpathians. In selected sympatric populations, detailed cytotype distribution was displayed to assess cytotype microhabitat preferences.

Plant morphology is affected by both cytotype and environment, and the environment may also relate to cytotype distribution. The evidence for potential synergies was explored using a causal network of piecewise structural equation models (SEMs; Shipley, 2009). To gain more insight, the environmental data were split into two subsets specifying local habitat conditions and climate (Supplementary Data Table S2), and the morphological data were split into three subsets consisting of variables characterizing the morphology of the leaf, flower and shoot (Supplemetary Material S2). These multivariate datasets were standardized and subjected to PCA to reduce their dimensionality. The first components for each subset accounted for 24–75 % of the variance in the datasets and were used in the SEM to represent original multivariate data. The variables were arranged in a directed acyclic graph, and a series of regression models was fitted to integrate plant morphology with ploidy levels and environment. Linear mixed models (LMMs; Pinheiro and Bates, 2000) were used to test the effect of cytotype and environment on plant morphology while accounting for multiple plants measured at the same sites. The structure of the LMMs involved random intercepts of sites and fixed effects of cytotype, habitat and climate. The models were screened for normality, homoscedasticity and spatial autocorrelation using standard diagnostic plots of residuals and spline correlograms (Bjørnstad and Falck, 2001). The LMMs for leaf and shoot morphology showed heterogeneous error variances, and thus the original models were reformulated by including an exponential variance function structure to fix the heteroscedasticity. No other violations of model assumptions were detected. To investigate the influence of the environment on the distribution of cytotypes, we fitted the ordinal outcomes of ploidy levels by using a cumulative logit mixed model with the Gauss–Hermite quadrature approximation (Tutz and Hennevogl, 1996). Likelihood-ratio tests were used to assess the overall significance of the models and the significance of individual terms. Marginal determination coefficients (R²_m) were calculated to quantify the proportion of the total variance explained by the models (Nakagawa et al., 2017). The relative contribution of individual variables was assessed using semi partial marginal determination coefficients (sR²_m) derived from commonality analysis (Ray‐Mukherjee et al., 2014). Correlative paths in the SEM were estimated using the Pearson correlation coefficient.

All analyses were performed in R language version 3.6.0 (R Core Team, 2019) using the libraries ellipse (Murdoch and Chow, 2018), ggplot2 (Wickham, 2016), lme (Pinheiro et al., 2019), ordinal (Christensen, 2019), ncf (Bjørnstad, 2019), performance (Lüdecke et al., 2020), pls (Mevik et al., 2019) and vegan (Oksanen et al., 2019).

RESULTS

Cytotype spatial distribution, mixed-ploidy populations and potential DNA aneuploids

A survey of 6011 individuals from 231 localities in Central Europe analysed by flow cytometry revealed three main ploidy levels (2x, 4x, 6x; Fig. 1, Supplementary Data Table S1), one odd ploidy level (5x in the locality Dreveník; site ID 14, Supplementary Data Material S3 and Supplementary Data Fig. S3) and several irregularities in the relative fluorescence suggesting differences in relative genome size and interpreted as potential DNA aneuploids (for all three ploidy levels from 17 localities; six, three and eight localities for 2x, 4x and 6x potential DNA aneuploids, respectively; Supplementary Data Material S3). Potential DNA-aneuploids were detected in samples where up to 10 plants were pooled, with the CV varying from 1.18 to 2.14 %.

Cytotypes were spatially clustered with parapatric distribution (Fig. 1, Supplementary Data Material S1). Hexaploids exclusively dominate the Pannonian Basin with a warm and continental climate and the westernmost part of the Western Carpathians (site IDs: 32, 33). Hexaploids share a long contact zone with tetraploids at the borders between the Western Carpathians and the Pannonian Basin, and another large area of tetraploids was found in the Bohemian Massif (i.e. Southern and Central Bohemia). Additional populations of tetraploids were scattered in South Moravia where the Bohemian Massif, the Pannonian Basin and the Western Carpathians are in contact (site IDs: 57, 66, 90, 97, 101) and in relict-like habitats such as serpentinite outcrops and rocky river valleys within the diploid range (Western and Central Bohemia: site IDs: 128, 129, 139, 273, 274, 469) and the hexaploid range (Southern Slovakia, site ID: 30). Diploids are widespread but almost exclusively in the Bohemian Massif, where they occupy a large spectrum of habitats (ruderal roadsides, pastures, meadows, forest edges, rocks and castle ruins).

A comparison of our flow cytometry data with previously published chromosome counts from the same locality showed discrepancies in ploidy level in 38 % of populations (Supplementary Data Material S1; Gadella, 1964; Kovanda, 1967, 1970a, 1977, 1983, 2002), including mixed-ploidy populations (Kovanda, 1967, 1970a). All revised mixed-ploidy populations were found to be of uniform ploidy, with only the higher ploidy present (Supplementary Data Material S1). However, eight mixed-ploidy populations (2x–4x, 4x–5x–6x, 2x–6x) were detected elsewhere, with higher ploidies always a minority except for the 2x–6x mixed-ploidy population with an almost equal proportion of 2x and 6x (Fig. 1, Supplementary Data Table S1 and Fig. S3).

Environmental niche breadth and niche divergence among cytotypes

OPLS-DA revealed significant shifts in environmental niches among the cytotypes (Table 1). The cytotypes were separated along with the first predictive component that significantly accounted for almost 52 % of the variability in the dataset (Fig. 2A). The second component was statistically non-significant. The first discriminant function represents the main climatic gradient, the contrast between calcareous and siliceous bedrocks and, to some extent, the intensity of human pressure (Fig. 3; Supplementary Data Table S3). Hexaploids prefer hotter and drier habitats with calcareous bedrock, and diploids occupy sites with lower temperatures and higher humidity often situated on siliceous bedrock with higher anthropic impact. Tetraploids typically dwell in intermediate conditions, which is also apparent from their central position in the ordination space of the OPLS-DA (Fig. 2A). Considering pairwise differences, the environmental niche of hexaploids differed significantly from both diploids (variance explained by the first component = 57.7 %, P = 0.027) and tetraploids (61.2 %, P = 0.002). The niches of diploids and tetraploids overlapped considerably and were statistically indistinguishable, either between parapatric or between allopatric cytotype pairs (Table 1).

Table 1.

Summary of the tests for differences in environmental niche shifts, environmental niche breadth, morphology and morphological variability among three cytotypes (2x, 4x and 6x) of Campanula rotundifolia agg. in Central Europe (tetraploids are divided into two groups based on their geographical occurrence: Western Carpathians = WC, Bohemian Massif = BM). Tests of niche shifts and morphological differences are based on OPLS-DA, and the table shows variance explained by the predictive components (comp 1, comp 2) along with probabilities in parentheses. Tests of differences in niche breadth and morphological variability are based on the homogeneity of multivariate dispersion, and the table shows F statistics (F) and respective probabilities in parentheses. Results significant at α = 5 % are highlighted in bold. Pairwise comparisons of the significant results are given in the contrast column. The type of cytotype spatial isolation (parapatry, allopatry) is indicated for pairwise comparisons.

Model	Contact zone	Environmental niche shift			Environmental niche breadth		Morphological differences			Morphological variability
		Comp 1 (%)	Comp 2 (%)	Contrast	F	Contrast	Comp 1 (%)	Comp 2 (%)	Contrast	F	Contrast
All (2x vs. 4x vs. 6x)		52 (0.027)	15 (0.517)	2x ≠ 6x, 4x ≠ 6x	11.2 (< 0.001)	6x < 2x < 4x	22 (< 0.001)	15 (0.181)	2x ≠ 6x, 2x ≠ 4x	0.4 (0.691)	–
2x vs. 4x BM	para	35 (0.475)	–	–	< 0.1 (0.888)	–	14 (0.331)	–	–	0.7 (0.417)	–
2x vs. 4x WC	allo	31 (0.555)	–	–	5.7 (0.018)	2x < 4x WC	22 (< 0.001)	–	2x ≠ 4x WC	0.8 (0.370)	–
6x vs. 4x BM	allo	52 (0.029)	–	6x ≠ 4x BM	11.5 (0.001)	6x < 4x BM	20 (0.014)	–	6x ≠ 4x BM	1.2 (0.287)	–
6x vs. 4x WC	para	65 (0.003)	–	6x ≠ 4x WC	14.6 (< 0.001)	6x < 4x WC	11 (0.466)	–	–	0.6 (0.457)	–
2x vs. 6x	para	58 (0.027)	–	2x ≠ 6x	10.2 (0.002)	6x < 2x	26 (< 0.001)	–	2x ≠ 6x	0.3 (0.577)	–
4x WC vs. 4x BM	allo	31 (0.485)	–	–	2.7 (0.108)	–	19 (0.012)	–	4x WC ≠ 4x BM	1.9 (0.174)	–

Fig. 2.

OPLS-DA score plots of predictive components with 95 % confidence ellipses showing differences in environmental niches (A) and morphology (B) among three major cytotypes of Campanula rotundifolia agg. The relative proportion of variance explained by the components is displayed in parentheses along with probabilities (P). Cross-validated classification accuracies and sensitivities are given as percentages for each model and cytotype, respectively.

Fig. 3.

Variable weights for the first (significant) predictive components of OPLS-DAs (Fig. 2) performed on environmental conditions (A) and morphological characteristics (B) of Campanula rotundifolia agg. cytotypes. For abbreviations of variable names, see Supplementary Data Material S2 and Table S3.

The environmental niche breadth differed significantly (F = 11.2, P < 0.001), and the environmental space occupied by the cytotypes decreased not gradually but in the following direction: 4x > 2x > 6x (Table 1). The same trend was recorded for parapatric and allopatric pairwise comparisons, although niche breadth between parapatric 2x–4x in the Bohemian Massif and the allopatric tetraploids did not differ significantly.

Morphological differences and variation among cytotypes

While cytotype morphological variation did not increase with increasing ploidy levels, their morphology differed significantly (Table 1). The cytotypes were gradually separated along with the first OPLS-DA component, which significantly accounted for more than 22 % of the variability in the morphological dataset (Figs 2B and 3; Supplementary Data Table S4). Diploid plants differed significantly from both tetraploids (variance explained by the first component = 19.9 %, P < 0.001) and hexaploids (25.7 %, P < 0.001) in vegetative characters. The latter cytotypes also differed, mainly in generative characters, although marginally non-significantly (14.5 %, P = 0.072). The density of the papillae on the ovary, the ratio of the length/width of the leaves in the middle and upper parts of the stem, the length of the leaves in the middle part of the stem, and other characteristics continually increased from diploids to hexaploids, while the width of the leaves in the middle and upper parts of the stem and the filament length continually decreased (Fig. 3; Supplementary Data Table S4). Comparisons of parapatric and allopatric cytotype pairs generally corroborated these results; however, two specific findings emerged (Table 1; Supplementary Data Table S4). First, allopatric tetraploids differed significantly in generative characters (shorter filaments and anthers and smaller ratios of the corolla length to calyx or corolla lobe lengths). Second, there were no differences between parapatric 2x–4x in the Bohemian Massif and 6x–4x from the Western Carpathians.

Environmental and ploidy level effects on morphology

Structural equation modelling showed that cytotype distribution was influenced by local habitat conditions (e.g. bedrock, PAR and altitude) rather than climatic characteristics, although both groups of environmental features were positively correlated (Fig. 4). Ploidy level significantly affects the shape of leaves (F = 22.1, P < 0.001), flowers (F = 19.3, P < 0.001) and, to a lesser extent, the overall plant (F = 6.1, P = 0.002). Environmental conditions significantly influenced leaf morphology (habitat effect: F = 39.4, P < 0.001; climate effect: F = 10.9, P = 0.001) and shoot morphology (climate effect: F = 7.5, P = 0.007), while the shape of generative organs was influenced only by ploidy level.

Fig. 4.

Structural equation model linking environmental characteristics (habitat and climate) with the distribution of cytotypes, leaf morphology, flower morphology and shoot morphology. Arrows represent significant (solid lines) and non-significant (dashed lines) relationships between variables. Unidirectional arrows denote direct effects of one variable on another, and their size is proportional to semi-partial marginal determination coefficients (sR²). Bidirectional arrows denote parwise correlations between variables, and their size is proportional to Pearson correlation coefficients (r). For each response variable, variance explained by predictors (marginal determination coefficients – R²) and statistical significance of whole models (P) are displayed. Morphological and environmental characteristics are latent variables derived as first components from principal component analyses on the original data. The most important original variables (rectangles) are shown along with their loadings on the latent morphological and environmental variables. For abbreviations of variable names, see Supplementary Data Material S2 and Table S3.

DISCUSSION

The current cytotype distribution

Central Europe is acknowledged as a potential region of origin of Campanula rotundifolia agg. (Sutherland et al., 2018; Wilson et al., 2020) because it may host the original diploids. From here, further spread of the complex was facilitated by polyploidization (Sutherland and Galloway, 2018). However, the longitudinal pattern of a ploidy increase from Central Europe to Western Europe and North America (Sutherland et al., 2018) is not supported in this study. Available chromosome counts (Böcher, 1936; Podlech, 1965) limited knowledge of the full diploid distribution, and the presence of polyploids in Central Europe was omitted, partly due to the intricate taxonomy of this group. The presented data favour the hypothesis of the polytopic origin of polyploids on a small (e.g. tetraploids from the Western Carpathians, the Bohemian Massif and autotetraploids in mixed-ploidy populations, Fig. 1) as well as a large geographical scale [e.g. hexaploids in the Pannonian Basin; Fig. 1, in Northern Italy (Fenaroli et al., 2013) and in Britain and Ireland (Wilson et al., 2020), tetraploids in the Western Carpathians/the Bohemian Massif; Fig. 1, in Britain (Wilson et al., 2020) and North America (Sutherland et al., 2018)].

In contrast to previous cytotype distributions based on published chromosome counts, cytotypes are geographically clustered (Fig. 1, Supplementary Data Material S1). Diploids have a distribution centre in the Bohemian Massif and are otherwise rare or endemic in Europe and North America, except for North Scandinavia (Laane et al., 1983). The largest discordance was found for the tetraploid cytotype detected in all regions (the Bohemian Massif, the Pannonian Basin and the Western Carpathians). Populations from the Pannonian Basin (populations with site IDs: 57, 66, 90, 97, 101) were not studied for morphological and environmental differentiation because of their low number and uncertain origin. These populations nested in the contact zone of the diploid and hexaploid cytotypes may represent autopolyploids that successfully established uniform ploidy populations. Alternatively, these populations may have remained from a formerly large tetraploid area that was now found to be hexaploid. However, none of these populations was included in studies involving chromosome counts (Supplementary Data Material S1). These tetraploids could also represent hybrids between diploids and hexaploids that may rarely be formed (Meeus et al., 2020) despite the relatively high postzygotic reproductive barrier between 2x and 6x cytotypes (Scott et al., 1998). Additional tetraploid populations were scattered within the diploid area, mostly on serpentinite outcrops (Fig. 1). The stressful nature of serpentinite environments (Brady et al., 2005) may lead to higher production of unreduced gametes (Ramsey and Schemske, 2002; Wilson et al., 2020) or may serve as a refugium for rare cytotypes, thus bypassing minority-cytotype exclusion (Kolář et al., 2012). A range expansion was detected for the hexaploid cytotype. It is almost exclusive in the Pannonian Basin, where hexaploids were mostly reported either from mixed-ploidy or formerly tetraploid populations (Gadella, 1964; Kovanda, 1967, 1970a, 1977, 1983; Mráz, 2005; Supplementary Data Material S1). Contact zones similar to those between 2x–6x and 6x–4x that roughly follow the borders of the studied regional features or between the Czech Republic and Slovakia were also detected for other ploidy-variable taxa (Mráz et al., 2008; Trávníček et al., 2010; Kobrlová et al., 2016; Macková et al., 2020).

Mixed-ploidy population dynamics and potential DNA aneuploids

Flow cytometry detected a shift in the frequency and location of mixed-ploidy populations (Supplementary Data Material S1). This approach screened the actual ploidy variation in the population, while previous studies used chromosome counts on seedlings from seeds collected at the locality (Gadella, 1964; Kovanda, 1966, 1970a, b; Rauchová, 2007). The possible establishment and survival of such seedlings may differ in the field and experimental cultivation (Sutherland and Galloway, 2016; Meeus et al., 2020). Experimental 2x–4x crosses yielded an almost equal frequency of 3x and 4x offspring (Sutherland and Galloway, 2016) However, in simulated 2x–4x open-pollinated contact zones, the ratio of homoploid and heteroploid crosses varied depending on cytotype frequency and pollinator preferences for the rare cytotype (Sutherland et al., 2020). Neither 3x nor 4x hybrids are likely to contribute to the interploidy gene flow in mixed-ploidy populations (Sutherland and Galloway, 2021). In simulated 2x–4x contact zones, the persistence of minority cytotypes was facilitated by pollinator preferences for the rare ploidy level and near-complete postzygotic reproductive isolation. The lack of intermediate cytotypes in mixed-ploidy populations (2x–4x, 2x–6x) or the 2x–4x contact zone supports the hypotheses of a strong triploid block and no secondary contact zone (Scott et al., 1998; Sutherland and Galloway, 2016; Lafon-Placette et al., 2017; Castro et al., 2020) where cytotype spatial separation may be enhanced via reinforcement (Sutherland and Galloway, 2021). Mixed-ploidy (2x–4x) localities within the diploid range thus probably represent a primary contact zone of 2x and auto-4x, mediated by a high production of diploid unreduced gametes (Castro et al., 2018) and low viability of possibly formed triploids. The lack of tetraploids, pentaploids or other cytotypes in the 2x–6x mixed-ploidy population suggests strong endosperm barriers between these two cytotypes (Scott et al., 1998; Meeus et al., 2020), supported by altitudinal cytotype separation (Schönswetter et al., 2007a). An almost similar proportion of diploids and hexaploids suggests it is a secondary contact zone. Unlike 2x–4x experimental crosses of Campanula rotundifolia agg. that yielded relatively balanced frequencies of 3x and 4x offspring, the 4x–6x crosses yielded almost exclusively pentaploids (Sutherland and Galloway, 2016). However, in simulated 4x–6x open-pollinated contact zones, the pollinator preferences for the rare cytotype altered the ratio of homoploid and heteroploid crosses with regard to cytotype frequency (Sutherland et al., 2020). In contrast to simulated 2x–4x contact zones, in simulated 4x–6x contact zones, the lack of postzygotic reproductive barriers and frequent formation of viable and fertile 5x may lead to asymmetric gene flow depending on cytotype frequency and introgression (Sutherland et al., 2020; Sutherland and Galloway, 2021). In contrast to the results of Sutherland et al. (2020), Wilson et al. (2020) detected a variable one-sided barrier favouring the persistence of tetraploids in 4x–6x mixed-ploidy populations in Britain. We found pentaploids and hexaploids within a predominantly tetraploid population (Fig. S3 and Supplementary Data Material S3), similar to the pattern in Britain (Wilson et al., 2020), Germany (Nierbauer et al., 2017) and North America (Sutherland and Galloway, 2018). These hexaploids are probably autopolyploids and form pentaploids via a 4x–6x backcross, and pentaploids may further participate in the heteroploid hybridization (Laport et al., 2016; Wilson et al., 2020; Peskoller et al., 2021; Sutherland and Galloway, 2021). At this locality, higher ploidies were found near tourist paths, which implies a role of human-induced stress in the higher production of unreduced gametes (Wilson et al., 2020).

Potential DNA aneuploids were found in all cytotypes, including pentaploids (Supplementary Data Material S3). Chromosome counts for 4x C. gentilis with 68 euploid chromosomes detected a variation of 67, 70, 73 and 75 chromosomes for aneuploids with a respective deviation of the DAPI ratio in flow cytometry analyses (Rauchová, 2007). Similar deviations were observed in Britain at sites with heavy metal soil for tetraploids and autohexaploids (Wilson et al., 2020). This finding further supports the role of environmental stress in the production of unreduced gametes (Ramsey and Schemske, 1998). However, Rauchová (2007) also counted an individual with a DAPI ratio corresponding to a potential DNA aneuploid as a tetraploid with 68 euploid chromosomes, which suggests intraspecific genome size variation. A mix of potential DNA aneuploids and individuals with different genome sizes was also observed for the studied individuals (Supplementary Data Material S3).

A shift in cytotype distribution

The shift between published and observed ploidy distribution favouring higher ploidy levels in mixed- and uniform-ploidy populations (Supplementary Data Material S1) suggests that polyploids are well established and have higher colonization success. However, the time needed to successfully replace the lower ploidy detected by earlier chromosome counting would be relatively short (M. Kovanda performed his studies on Campanula rotundifolia agg. 50 years ago), and in simulated contact zones of Campanula rotundifolia agg., the cytotypes produced progeny of the same ploidy level more frequently than expected under random mating (Sutherland et al., 2020; Wilson et al., 2020). Thus, the likelihood of counting a seedling of a different ploidy than that of the mother plant is relatively low, especially in 4x or 4x–6x populations. Moreover, Wilson et al. (2020) showed that under open pollination, tetraploid mothers produced almost exclusively tetraploid offspring, while hexaploid progeny was more variable, including many pentaploids and aneuploids.

Although our sampling for the ploidy level estimates by flow cytometry did not cover the entire population, two cytotypes were often detected in populations from which only 11 individuals were sampled (Supplementary Data Table S1). Localities with published chromosome counts were sampled more thoroughly (30 individuals at a minimum; Supplementary Data Material S1). Nevertheless, the previously published cytotypes still may have been undetected in some undersampled populations.

Several studies on other plant species also show large discrepancies between published and observed ploidy variation in a similar pattern (exclusive occurrence of higher ploidy levels) for populations counted by M. Kovanda, and they assigned these differences to errors or difficulties in chromosome counting or to sampling artefacts (Laane et al., 1983; Weiss et al., 2002; Rauchová, 2007; Vít et al., 2012; Macková et al., 2020; Wilson et al., 2020). This raises awareness that the original data do not reflect the actual cytogeography. However, these data are still kept in online karyological databases (e.g. Marhold et al., 2007: database available at http://www.chromosomes.sav.sk; Rice et al., 2015: database available at http://ccdb.tau.ac.il), which may bias further metanalyses using these sources (Porturas et al., 2019; Rice et al., 2019). Therefore, the use of published data without verifying the cytotype distribution in the field with high-throughput methods, such as flow cytometry, is highly questionable. More cytogeographical studies are thus needed because they very often discover unexpected variability, either at the local (Chumová et al., 2015; Trávníček et al., 2012; Caperta et al., 2017; Nierbauer et al., 2017; Wilson et al., 2020) or at broad geographical scales (Krejčíková et al., 2013; Čertner et al., 2017; Nierbauer et al., 2017; Paule et al., 2017; Hanušová et al., 2019; Rejlová et al., 2019). New data on ploidy variation provide valuable insights into the evolutionary mechanisms shaping the gene flow dynamics of polyploids and, consequently, speciation. Minority cytotypes can help to detect interploidy hybridization through a triploid or a pentaploid bridge (Mandáková et al., 2013; Peskoller et al., 2021) with consequences for conservation genetics (Nierbauer et al., 2017; Macková et al., 2018), to detect the dynamic formation of unreduced gametes and neopolyploids (Wilson et al., 2020; Sutherland and Galloway, 2021) and reveal cryptic variation that may result in speciation events (Flatscher et al., 2015). Describing cytotypes as different species has its pitfalls such as in the case of Campanula gentilis and C. moravica. These two taxa were delimited from C. rotundifolia agg. mainly because of their different chromosome numbers (C. moravica) or the presence of mixed-ploidy populations (C. gentilis), and minute morphological differences related to WGD. The morphological differentiation was later doubted (Kovanda, 2002; Rauchová, 2007; Šemberová, 2013), leaving geographical distribution and chromosome number the only reliable identifiers of C. gentilis and C. moravica, respectively. A shift in cytotype distribution is also of relevance to the loci classici of C. gentilis and C. moravica, which implies a need for a detailed taxonomical revision. Taxonomic complexity, unclear geographical distribution and low morphological differentiation of individual species led us to omit the species names in this study.

Cytotype morphological and environmental differentiation as prezygotic reproductive barriers

Higher morphological variation, often created by polyploidy, was not confirmed in higher polyploids (6x) despite the trend toward larger size observed in some traits, especially leaf length and width. Similarly, morphological differentiation was more pronounced between diploids and polyploids than between higher ploidies (Table 1, Fig. 2B; Supplementary Data Table S4; Porturas et al., 2019). Only cytotype pairs involving hexaploids and those that were spatially isolated differed morphologically. Morphological changes induced by WGD may vary within the same cytotype (Laport and Ramsey, 2015; López-Jurado et al., 2019), suggesting that allopatric tetraploids probably represent lineages independently formed via recurrent formation. The lack of an environmental niche shift suggests that adaptations to local microclimatic or microhabitat conditions mirrored by the different geomorphological histories and flora of the two regions (Kaplan, 2012; Mráz and Ronikier, 2016) played a role in the tetraploid morphological differentiation (Castro et al., 2018; López-Jurado et al., 2019; Wilson et al., 2020). The lower postzygotic reproductive barrier and higher gene flow between higher ploidy levels (Sutherland and Galloway, 2021) would need to be compensated for by a strong prezygotic reproductive isolation (e.g. by altering phenology or pollinator preferences). However, in simulated contact zones of C. rotundifolia agg., pollinators did not prefer a specific cytotype but they overvisited the rare one (Sutherland et al., 2020). This contrasts with studies comparing plant–pollinator interactions along an elevational gradient where differences in plant morphology correlated with altitude and a related shift in pollinator size (Maad et al., 2013). In addition, differences in phenology were detected between plants from different European countries (Preite et al., 2015) but not on a smaller regional scale, among the plants from Britain and Ireland (Wilson et al., 2020). While diploids from across Europe flowered later than some European tetraploids (Gadella, 1964), diploids and tetraploids in the Bohemian Massif started flowering earlier than hexaploids and tetraploids in the Pannonian Basin and Western Carpathians, suggesting a longitudinal shift in phenology.

The narrowest environmental niche breadth and the niche shift from all other cytotypes suggest niche specialization of the hexaploid cytotype and support the niche-filling hypothesis (López-Jurado et al., 2019). In contrast, tetraploids had the widest niche, similar to the pattern of niche expansion observed in other polyploid complexes (Karunarathne et al., 2018; López-Jurado et al., 2019; Molina-Henao and Hopkins, 2019). However, this was true only when tetraploids from the Bohemian Massif and the Western Carpathians were merged in the analysis. The environmental niche breadth of tetraploids from each area (either Bohemian Massif or Western Carpathians) was still broader than that of hexaploids but narrower than that of diploids (Table 1). Diploids had the broadest niche. While polyploids may benefit from gene neofunctionalization serving as pre-adaptation to a broader distribution, diploids, as the original and first cytotype, had the longest time to spread and adapt to the local environment.

The lack of niche shift (Table 1) between diploids and tetraploids, either in parapatry or in allopatry, may indicate a partly sufficient postzygotic reproductive barrier because no triploids were detected in the field. The triploid block may be bypassed by the tetraploid progeny (Sutherland and Galloway, 2016), and ongoing gene flow could explain the lack of morphological differentiation. However, no gene flow (Sutherland and Galloway, 2021) and no mixed-ploidy populations were found in the contact zone between diploids and tetraploids in the Bohemian Massif (Fig. 1), suggesting low viability of triploid and tetraploid hybrids. In contrast, the lack of pentaploid hybrids between parapatric tetraploids and hexaploids despite the lack of morphological differentiation may imply that the environmental niche shift is an efficient prezygotic reproductive barrier promoting assortative mating (Husband and Schemske, 2000; Sonnleitner et al., 2010; Castro et al., 2020). However, the environmental niche shift turned out not to be affected by the level of cytotype spatial isolation or strength of the postzygotic reproductive isolation because it was detected only for cytotype pairs, including hexaploids, irrespective of parapatry or allopatry. Thus, it is probably not due to reinforcement but rather reflects the niche specialization of the hexaploid cytotype according to the niche filling hypothesis (López-Jurado et al., 2019).

The strong postzygotic reproductive isolation in 2x–4x contact zones results in neotetraploids and unfit triploids not participating in heteroploid gene flow. Isolated cytotypes thus may, via reinforcement or minority cytotype exclusion, undergo independent evolution (Sutherland and Galloway, 2021). In contrast, lower postzygotic isolation of polyploids allows the formation of viable and fertile pentaploids participating in 4x–6x gene flow (Wilson et al., 2020; Sutherland and Galloway, 2021). Even if slowing the divergence (Sutherland and Galloway, 2021), pentaploids may represent an important bridge between the two cytotypes, allowing polyploids to benefit from a larger shared gene pool. We did not detect morphological or phenological differentiation between parapatric polyploids (4x Western Carpathians –6x) that could imply reinforcement. Despite the lack of selection for assortative mating, no mixed-ploidy populations or pentaploids were detected in the contact zone. Environmental niche shifts driven by the specialization of the hexaploids may thus serve as an efficient prezygotic reproductive barrier.

Consequences of climate, environment and ploidy on plant morphology

WGD immediately manifests in plant phenotype (Laport and Ramsey, 2015), phenology (Simón-Porcar et al., 2017) and reproductive mode (Pannell et al., 2004; Meeus et al, 2020). However, reproductive isolation is more pronounced in established polyploids than in synthetic neopolyploids (Husband et al., 2016; Porturas et al., 2019), which suggests that local adaptation, environmental niche shifts (López-Jurado et al., 2019) and reinforcement enhancing assortative mating (Kirkpatrick, 2000; Hopkins, 2013) could act as prezygotic reproductive barriers. The interactions between climate, environment and ploidy level on cytotype morphology and distribution are complex (Fig. 4). Despite niche specialization of hexaploids, the cytotype distribution was more influenced by local habitat conditions (e.g. bedrock, PAR and altitude) than climate. Leaf and shoot morphology were influenced by ploidy and environmental conditions, which mirrors the intricate taxonomic complexity and endemism of Campanula rotundifolia agg. (Podlech, 1965; Kovanda, 2002; Kovačić, 2004; Mansion et al., 2012). Concerning local adaptations, the morphology of leaves and the overall plant may provide the best adaptive potential (Laport and Ramsey, 2015) without affecting pollinators or thus plant reproduction. Therefore, despite the tendency towards larger size in some traits, the longer and narrower leaves of hexaploids may be a xeromorphic adaptation to the drier and warmer Pannonian Basin, while diploids with wider and shorter leaves prefer the humid and colder Bohemian Massif (Figs 1 and 3). In contrast, generative traits were influenced by ploidy level only and differed among the higher polyploids (Fig. 3). Through floral morphology, a shift in pollinator preferences or phenology may imply reinforcement and serve as one of the prezygotic reproductive barriers (Hopkins, 2013; Laport and Ramsey, 2015), although this is not immediately apparent (Castro et al., 2020).

CONCLUSION

The discrepancies revealed between published and revised ploidy variation imply a need for high-throughput methods complementing chromosome counting. Environmental niche specialization of hexaploids of Campanula rotundifolia agg. may ensure their reproductive isolation from parapatric tetraploids under the lack of a strong postzygotic reproductive barrier (Sutherland and Galloway, 2016). In contrast, morphological differentiation was detected almost exclusively for spatially isolated cytotypes, including allopatric tetraploids, which implies their independent origin (Wei et al., 2017; López-Jurado et al., 2019).

The absence of hybrids despite the lack of morphological and environmental differentiation between parapatric diploids and tetraploids emphasizes the strength of the triploid block. However, ongoing gene flow towards tetraploids potentially occurs due to the formation of tetraploids via unreduced gametes (Sutherland and Galloway, 2016), which could explain the lack of morphological differentiation between parapatric diploids and tetraploids.

Cytotype distribution was influenced by local habitat conditions, and the level of cytotype spatial isolation had no impact on reproductive isolation. The triploid block and environmental niche shifts are the main mechanisms of reproductive isolation, driven both by the ploidy level and environment. Polyploidy is the only variable influencing flower shape, which emphasizes the role of WGD as a prezygotic reproductive barrier (Kennedy et al., 2006). The lack of pollinator fidelity between newly in silico created polyploids and their diploid ancestors (Porturas et al., 2019) and between evolutionarily young cytotypes of Campanula rotundifolia agg. (Roquet et al., 2009; Sutherland et al., 2020) implies a need for longer adaptive coevolution for pollinators to effectively distinguish between cytotypes. The synergy between polyploidy, environment and morphology is complex and varies between different ploidy levels and their level of spatial isolation.

SUPPLEMENTARY DATA

Supplementary data are available online at https://academic.oup.com/aob and consist of the following. Material S1: FCM estimated ploidy level in localities with known chromosome counts. Material S2: Morphological variables. Material S3: Relative genome size and potential DNA aneuploids. Figure S1: Photos of Campanula rotundifolia agg. Figure S2: Validation plots for the OPLS-DA. Figure S3: A mixed-ploidy population. Table S1: Localities. Table S2: Environmental variables. Table S3: Summary of environmental variables. Table S4: Summary of morphological variables.

FUNDING

The study was supported by Charles University [project GA UK No. 670213] to KŠ. KŠ and RS were supported by the long-term research project of the Czech Academy of Sciences, Institute of Botany (RVO 67985939). MS was supported by the Operational Programme Integrated Infrastructure (OPII), funded by the ERDF [ITMS 313011T721].

CONFLICT OF INTEREST

The authors declare no conflict of interest.