Niche differences may explain the geographic distribution of cytotypes in Erysimum mediohispanicum

Study populations

We gathered information on ploidy level, floral phenotype and climate from 118 E. mediohispanicum populations, representing the entire distribution range of the species. To minimise the impact that environmental conditions of a particular year on floral phenotypes, we measured phenotypic traits for at least 2 years in most populations (see Table S1 for further details on location, ploidy, sample size and sampling strategy). Using leaf tissue from individuals of a single year per population, we estimated ploidy level of 52 populations by analysing microsatellite electropherograms obtained following Muñoz-Pajares et al. (2011). To confirm the accuracy of microsatellite markers to discriminate between diploid and polyploid individuals, we used flow cytometry (FCM) to re-estimate genome size of 16 populations previously studied using microsatellite markers. Because ploidy level was homogeneous in the five populations where FCM was applied to more than 20 individuals (representing 15–30% of the entire populations), ploidy levels of 51 additional populations were estimated using a low number of individuals (average of three individuals per population), except in areas of special interest (where we analysed an average of 17 individuals per population; Table S1). Finally, in 15 populations, ploidy level was obtained from the literature, being estimations based on chromosome counts (Polatschek 1979; Clot 1992). Details on methods to estimate ploidy level using both, FCM and microsatellite analyses, are provided in Appendix S1 and Fig. S1.

Determination of floral phenotype

We studied the floral phenotype of 54 populations, characterising at least 30 individuals per population. However, some populations were smaller; in those cases, all flowering individuals occurring in the population were studied (Table S1). In each plant we measured: (i) stalk height, height of the tallest flowering stalk, obtained using a measuring tape (error ± 0.5 cm); (ii) stalk diameter, diameter at the base of the tallest flowering stalk; (iii) flower number, counting all flowers and flower buds produced per plant; (iv) corolla diameter, distance between the apical edges of two opposite petals; (v) corolla tube length, distance between the corolla tube aperture and base of the sepals; (vi) corolla tube width, internal space between petals at the top of the corolla tube aperture, and estimated as the difference between corolla diameter minus length of two opposite petals. Traits (ii) and (iv) to (vi) were measured using digital callipers with 0.1 mm resolution. Traits (iv) to (vi) were estimated in one flower per plant.

Climate variables

Interpolated climate variables at 0.5 arc min resolution were obtained for the studied populations from the Worldclim 1.4 database (Hijmans et al. 2005) using the raster package in R (Hijmans 2016). Specifically, we selected three climatic variables based on their expected biological importance to E. mediohispanicum reproductive cycle: Precipitation (PR, mm·day⁻¹) during the period of flowering (and pollination) and fruiting; Mean temperature (TM, °C) during seedling establishment; and Temperature range (TR, °C) during the final vegetative growth before plant reproduction. To better represent climatic environment, we also retained the bioclimatic variables showing Pearson's correlation coefficients lower than 0.7 (Table S2), raising the total number of climatic variables used to seven: PR, TM, TR, Bio3 (isothermality, that is, the quotient between day-to-night and summer-to-winter temperature oscillation), Bio8 (mean temperature of wettest quarter), Bio9 (mean temperature of driest quarter), and Bio19 (precipitation of coldest quarter). See Appendix S1 for a complete description on the variables used in this work.

Correlations between phenotype, ploidy and climate

Prior to any analyses, we log-transformed phenotypic traits departing from normality (namely, stalk diameter, stalk height and number of flowers). The relationships between phenotypic traits, ploidy level and climate variables were evaluated using canonical correlation analyses (CCA). We normalised all climatic and phenotypic variables and computed the CCA using the CCA package in R (González & Déjean 2012).

Environmental niche modelling and niche overlap

To obtain a more accurate estimate of the environmental niche of the two cytotypes, we also downloaded six variables describing soil properties from the European Soil Database version 2 Raster Library (Panagos 2006). Specifically, we used parmado (dominant parent material), wrbadj1 (first soil adjective code), DR (depth to rock), oc_top (topsoil organic carbon content), text (dominant surface textural class) and usedo (dominant land use). Finally, we used the same database to download the altitude layer to account for effects of population elevation. Because altitude had a strong negative correlation with TM (Pearson's coefficient = −0.87; Table S2), the latter was excluded for analyses including altitude.

Niche modelling was performed with maximum entropy modelling using the MaxEnt software (Phillips & Dudík 2008) with default parameters, except for number of replicates (15), percentage of random tests (25) and maximum number of iterations (5000). We used the area under the curve statistic (AUC) to evaluate model accuracy and the relative contribution of each variable to the final model. Overlap between niches of both E. mediohispanicum cytotypes was quantified using the Schoener's D (See Appendix S1 for further details).

Distribution pattern of cytotypes in E. mediohispanicum

Using FCM analysis, we estimated ploidy level in 416 individuals from 67 populations (Table S1) and found that 211 of them were diploid (2x; average genome size: 0.47 pg·2C⁻¹) and 200 were tetraploid (4x; average genome size: 0.99 pg·2C⁻¹). However, 2x and 4x individuals inhabit different populations of homogeneous ploidy (Fig. S2). In fact, only four out of the 63 studied populations contained more than one cytotype (Table 1). From these, only in one population (Em93) were 2x and 4x found growing in sympatry and, even in that case, their frequencies within the population were extremely unbalanced (2x + 4x + aneuploids; 27:1:1; Table S1). The remaining three populations with ploidy heterogeneity were Em54 (4x + 5x; 2:1; genome size of the 5x individual was 1.26 pg·2C⁻¹), Em71 (4x + aneuploids, 13:2; genome size of aneuploids was 1.17 and 1.18 pg·2C⁻¹) and Em98 (3x + 4x; 1:4, genome size of the 3x individual was 0.75 pg·2C⁻¹; Table S1, Fig. S2).

The geographic distribution of ploidies was mostly disjunct, with diploid individuals being restricted to the southernmost part of the Iberian Peninsula and tetraploids occurring in the central and northern parts of the Iberian Peninsula (Table 1; Fig. S2). Interestingly, a contact zone between diploids and tetraploids was found between these two main areas. Specifically, the contact zone is located in the Prebaetic Ranges of the Baetic Mountains, less than 50 km from the northernmost diploid populations and 100 km from the southernmost tetraploid populations (Fig. S2). Within the contact zone we did not find any populations showing both cytotypes in a similar proportion, but a single tetraploid individual was found in a diploid population; the two populations having aneuploid individuals were also detected in this region (Fig. S2, Table 1). The shortest distance between populations with different ploidies in the contact zone was 6.8 km.

Correlations between phenotype, ploidy and climate

Even though cytotypes could not be distinguished by phenotype in natural populations, after analysing 4733 individuals we found significant differences in several traits (Fig. 1). Specifically, flowering stalks were significantly wider in diploids than in tetraploids, while for stalk height the opposite trend was observed, with tetraploids being taller than diploids. The corolla tube width was also larger in tetraploids than in diploids, being the trait showing the largest differences. The remaining phenotypic traits were not statistically different between ploidy levels (Fig. 1). Similar results were obtained using only the subset of individuals for which both ploidy levels and phenotypic traits were measured (data not shown).

The correlations between ploidy and both stalk diameter and corolla tube width (negative and positive, respectively) were confirmed with CCA results (Fig. 2). The CCA clearly separated diploid and tetraploid populations in dimension 1 (Fig. 2A), which, in addition to ploidy level, corolla tube width and stalk diameter, importantly depends on precipitation and temperature (Fig. 2B, Table S3). Specifically, polyploid populations seem to occur in areas with higher precipitation during flowering and fruiting (PR) and mean temperature during the wettest quarter (Bio8), but lower precipitation of the coldest quarter (Bio19) and mean temperature of driest quarter (Bio9).

Environmental niche modelling and niche overlap

Niche models estimated using climate variables, altitude and soil variables had very high AUC scores for 2x (mean 0.985 ± 0.026; ±SD) and 4x (mean 0.928 ± 0.046) populations, suggesting negligible rates of false negative and false positive suitability predictions. In fact, most of the studied populations were in areas having high predicted probability (Fig. 3). Interestingly, the models point to the observed contact area as one of the few regions in the Iberian Peninsula where environment appears highly suitable for both cytotypes (Fig. 3). Niche overlap was 37.9%, with diploid and tetraploid niches overlapping significantly less than expected by chance according to the equivalency test (P = 0.01). However, according to the similarity test, niche overlap was not significantly lower than random (2N versus 4N, P = 0.57; 4N versus 2N, P = 0.56). These findings are compatible with the existence of subtle differences between the niches occupied by the two cytotypes. We obtained congruent results after using only climate (continuous) variables for niche models estimation (see Appendix S2).

Mean temperature during seedling establishment (TM), altitude and temperature range during bud development (TR) were the variables having the highest contribution to the final models for 4x populations, whereas precipitation during flowering and fruiting (PR), altitude and TR were the most important variables for 2x populations (Table S4). Specifically, 2x populations tend to occur at lower PR but higher TR and TM. Regarding soil variables, the dominant parent material (parmado) had the most consistent influence in both models. According to that variable, E. mediohispanicum mainly grows on limestone, as 60% of the studied populations were found in that type of soil. However, this value importantly varies with ploidy, being as high as 81% for 2x, but only 48% for 4x. Indeed, 4x individuals are found growing on calcareous sandstone in 24% of the populations, on marl soils in 12% of the cases and on other non-marl soils in 16% of the populations (Table S5).

Adaptive basis of ploidy

In E. mediohispanicum populations, the ploidy level is lower in the south than in the north of the Iberian Peninsula. Such a pattern has already been observed in other plant species in this geographic area, such as Arenaria tetraquetra (Vargas 2003). However, opposite (Brachypodium distachyon, Manzaneda et al. 2012; Mercurialis annua, Buggs & Pannell 2006) and more complex patterns (Cardamine pratensis, Lihová et al. 2003; Dianthus broteri, Balao et al. 2009) have also been found. This lack of a general pattern sustains the doubt as to whether the geographic distribution of cytotypes in the Iberian Peninsula is adaptive or, conversely, if it responds to other ecological processes (Lumaret et al. 1987; Buggs & Pannell 2006; Soltis et al. 2010). In the case of E. mediohispanicum, niches of the two ploidies were significantly different according to the equivalency test but not according to the similarity test. These results allowed us to reject the null hypothesis that both cytotypes have a highly conserved environmental niche, but rather suggest the existence of subtle differences between niches (Visger et al. 2016). However, despite niche differences not being strong enough to be detected by the low-sensitivity similarity test, subtle departures from niche equivalency may lead autopolyploids to escape from minimum cytotype exclusion (Visger et al. 2016). Climatic factors and soil properties seem to provide an adaptive basis to explain the distribution of E. mediohispanicum cytotypes. Indeed, our models identified several environmental variables consistently contributing to explain the disjunct distribution of cytotypes, namely altitude, soil dominant parent material and various climate variables.

Erysimum mediohispanicum diploid populations occur at higher elevations than tetraploids (mean elevations: 1536 and 1088 m a.s.l., respectively), with populations in the contact zone occurring at intermediate elevations (Fig. 4). This elevational segregation of diploid and tetraploid populations seems to be frequent in other diploid–polyploid complexes, such as Chamerion angustifolium (Husband & Schemske 1998), Lotus corniculatus (Gauthier et al. 1998), Anthoxanthum alpinum and A. odoratum (Flegrová & Krahulec 1999 and references therein), Taraxacum section Ruderalia (Meirmans et al. 2003) and Larrea tridentata (Laport et al. 2016). Differences in cytotype distribution associated with population elevation are considered difficult to separate from the effects, among other factors, of latitude, climate and soil (Flegrová & Krahulec 1999). Using niche modelling, we were able to discriminate the importance of elevation from other factors. However, evaluating fitness of cytotypes using reciprocal transplants is required to definitively confirm the adaptive basis of the association of cytotype to elevation (Husband & Schemske 1998; Flegrová & Krahulec 1999; Martin & Husband 2013).

Importantly, phenotypic traits and niche models depend on several climate variables. Interestingly, the three climate variables built on the basis of their putative influence on the E. mediohispanicum life cycle (i.e. precipitation during flowering and fruiting, mean temperature during seedling establishment and temperature range during bud development) were consistently selected as the most influential to explain cytotype distribution. This result demonstrates the importance of considering the biology of the study species when determining the appropriate variables, rather than merely using standard variables.

Our analyses suggest that tetraploid individuals inhabit areas with wetter summers (Fig. 4). This finding contrasts with the expectation that polyploids possess larger but fewer stomata, allowing them to reduce water loss through the leaves, and thus to inhabit dry regions (Levin 2002; te Beest et al. 2012; Manzaneda et al. 2012). Visger et al. (2016) recently found the same pattern in Tolmiea diplomenziesii (2x) and T. menziesii (4x), with a common garden experiment confirming that diploids made better use of water under drought conditions, potentially due to differences in xylem diameter of the two cytotypes. The same may occur in E. mediohispanicum, but specific physiological measurements are necessary to confirm this hypothesis.

After including the valuable information on soil properties in our environmental niche model reconstruction, we found that tetraploids occur on calcareous sandstones much more frequently than diploids, which mostly grow on limestone. Soil differences in water desorption (faster for limestone; Vázquez et al. 2013) may reinforce the lower availability of water in 2x populations during the driest months, thus contributing to a more drastic effect of summer drought. Another interesting result of soil comparison is that 4x populations grow in more diverse soil types, suggesting that they may have increased ecological amplitude. This result is congruent with previous studies on other plant species that attributed larger ecological amplitudes to polyploids due to their increased genomic flexibility and adaptive potential (Levin 2002; Parisod et al. 2010; Wallace et al. 2017).

Contact zone and gene flow between cytotypes

Niche differences between cytotypes seem to be maintained in the contact zone, with diploid and tetraploid populations growing at different altitudes, TM and PR (Fig. 4). This suggests that in the contact zone, cytotypes are distributed in a mosaic zone (Harrison & Rand 1989). In fact, differences in these climate variables are more dramatic at the fine spatial scale of the contact zone than after pooling the whole distribution range (Fig. 4). This result supports the association between ploidy level and environmental variables, suggesting that the observed distribution of E. mediohispanicum cytotypes is, at least partially, explained by the niche differentiation hypothesis.

The Prebaetic Range is the only area where populations of each cytotype are close enough to allow gene flow to occur between them. It was also in this region where the only mixed ploidy population was found. Nevertheless, only one tetraploid individual was found out of 29 plants analysed from this population. This tetraploid plant either originated within the population (through the fusion of unreduced gametes) or was able to reach it from nearby tetraploid populations (the distance to the closest tetraploid population is 15 km). Either way, if this plant has no competitive advantage in fitness in comparison with the more abundant diploids, it will be subjected to strong frequency-dependent selection (Levin 1975). In that scenario, the mating system of the species may play an important role in maintenance of such minor cytotype individuals through self-fertilisation (Rausch & Morgan 2005). However, for this to occur, a similar ability of both cytotypes to inhabit the same population would also be required, which seems unlikely considering the E. mediohispanicum mosaic zone. Thus, despite our results not providing incontrovertible evidence for rejecting the influence of reproductive exclusion and historical factors, the observed distribution of cytotypes seems better explained through niche differentiation.

There was an extremely low frequency of individuals showing intermediate ploidies in the contact zone (only three aneuploid individuals were found in two populations, with a complete absence of triploids). This suggests that hybridisation between cytotypes is rare in nature and points to the occurrence of meiotic abnormalities as the mechanism underlying the low frequency of minority cytotypes observed in natural populations. Meiotic abnormalities (i.e. production of gametes with unexpected ploidy) are common in some species (Brownfield & Köhler 2011; De Storme & Geelen 2013) and have been described in both cytotypes of E. mediohispanicum (Blanca et al. 1992; Clot 1992). These abnormalities may also explain the occurrence of 3x and 5x individuals in the tetraploid populations Em98 and Em54, respectively, as none of the surroundings populations showed the expected ploidy level (2x and 6x, respectively) that could produce, through hybridisation, the individuals with the observed ploidy level.

Phenotypic differences

Most of the evolutionary advantages attributed to polyploids are related to phenotypic changes, especially for traits related to ecological interactions (Segraves et al. 1999; Thompson et al. 2004; Kennedy et al. 2006; Münzbergová 2006; Arvanitis et al. 2007). In this work, we evaluated the existence of significant differences between traits related to survival (plant size) and reproduction (corolla size). In general, we found that tetraploid individuals tend to present thinner stalks than diploids but are taller and their flowers are bigger, with the corolla tube width having the largest difference among cytotypes. These findings are in accordance with patterns reported in other species, with polyploids usually being taller, more robust and producing larger flowers than their diploid counterparts (te Beest et al. 2012 and references therein). Phenotypic changes associated with ploidy have consequences on floral visitors, with polyploids attracting different pollinator assemblages than diploids (Taylor & Smith 1979; Segraves & Thompson 1999). Although an exhaustive comparison must be done with the two cytotypes in sympatry, preliminary analyses have shown marginal, non-significant, differences between pollinator assemblages in diploid and tetraploid disjunct populations of E. mediohspanicum (manova r² = 0.04, P = 0.07; Muñoz-Pajares 2013).