Svalbard is a High-Arctic archipelago situated between 74°–81°N and 10°–35°E. S. oppositifolia is likely the most widely distributed vascular plant in Svalbard, and is often dominant in the vegetation (Elvebakk 1994; Alsos et al. 2012). Samples of S. oppositifolia were collected from 15 locations, covering all three bioclimatic zones present in Svalbard—the middle-Arctic tundra zone, the northern Arctic tundra zone and the Arctic polar desert zone (Elvebakk 2005; Walker et al. 2005)—and from different vegetation types, to cover the intraspecific variation in ecology and growth form. The collection period was from mid-July to mid-August 2011. Usually, we had one sampling site within each location. In five locations, we had two sampling sites: one in a ridge habitat and one in a slope habitat (Table 1; Fig. 1). In total, leaf material from 195 individuals was collected for ploidy analysis. Most material was dried in silica gel, but some samples were analysed fresh. The growth forms were classified subjectively by the observer into cushion type (the individual plant formed a dense cushion), intermediate type (the individual plant formed dense, cushion-like parts, but also trailing branches) and prostrate type (the individual plant had only trailing parts). In the immediate vicinity of 169 of the 195 sampled individuals, environmental and vegetation data were collected (Table 1). The distance between sampled individuals was usually 10 m but in some cases less, the minimum distance was 5 m. Vouchers of all sampled individuals are stored at the University Centre in Svalbard.

Vascular plant species were recorded using the point frame intercept method (Bråthen & Agberg 2004) with a 0.25 m² frame and 25 equidistant points. In addition, cover percentages of surface types were estimated (bryophytes, lichens, soil biological crust, litter and bare ground). Vascular plant species that were not recorded with the intercept method, but that occurred within the frame, were noted.

Soil moisture and soil temperature were measured at 5 cm depth in each plot. The soil moisture and soil temperature at location 9, 10, 11,13 and 15 (Table 1) were measured with a HOBO Micro Station Data Logger H21–002 with 2/2 Temp/RH sensors (MicroDAQ.com, Ltd., Contoocook, NH, USA). For the other plots, a digital thermometer was used to measure soil temperature, and moisture was estimated by appearance and feel of the soil (e.g., Klocke & Fischbach 1984). The soil moisture measurements were later transformed to classes to enable comparability to the estimated soil moisture. Soil samples were taken at 0–5 cm depth to determine pH, conductivity and loss on ignition (LOI). These samples were stored at 4°C until further processing.

The soil samples from each sampling site were pooled and sieved with a 2-mm meshed sieve. From each of the pooled and sieved soil samples, we took sub-samples for pH, conductivity measurements and LOI determination. The soil used for pH and conductivity measurements was oven-dried at 31.8°C for 24 h. Thereafter, 10 g soil was mixed with 25 ml distilled water, shaken and left to settle for 12 h. Measurements were taken with a VWR sympHony SP90M5 with a pH electrode and a conductivity electrode (VWR International, Radnor, PA, USA).

For the LOI determination, we followed the procedures described by Dean (1974) and improved by Heiri et al. (2001). Three sub-samples from each sampling site were oven-dried for 24 h at 105°C, then oven-burned for 4 h at 550°C. We ensured that the weight of each soil sample replicate was similar, as Heiri et al. (2001) showed that LOI at 550°C is dependent on sample size. The percentage LOI₅₅₀ was calculated as LOI₅₅₀=((DW₁₀₅–DW₅₅₀)/DW₁₀₅)*100 (Heiri et al. 2001).

All ploidy analyses were completed by Plant Cytometry Services, AG Schijndel, The Netherlands, using the following protocol: each leaf sample was prepared by chopping 20–50 mg with a razor blade in an ice-cold buffer containing 5 mM HEPES, 10 mM magnesium sulphate heptahydrate, 50 mM potassium chloride, 0.2% Triton X-100, 0.1% dithiothreitol, 1% polyvinylpyrrolidone and 2 µg/ml DAPI; buffer pH was 8.0. Approximately 2 ml of the solution containing cells and tissue were passed through a nylon filter of 50-µm mesh size. Fluorescence was measured with a CyFlow ML flow cytometer (Partec GmbH, Münster, Germany). The flow cytometer was equipped with a high-pressure mercury lamp (Osram HBO 100 long life), heat protection filter KG-1, excitation filters UG-1 and BG-38, dichroic mirrors TK 420 and TK 560 and an emission filter GG 435. In addition, FlowMax software (version 2.4, Partec GmbH, Münster, Germany) was used.

The measured ploidy levels (diploid, triploid and tetraploid) were sorted according to the registered growth forms (cushion, intermediate and prostrate) and analysed for association in a contingency table applying the program PAST version 2.15 (Hammer et al. 2001). A contingency table was also used to test the association between habitats (ridge or slope) and ploidy level and growth form.

In addition, the results from the triploid counts were removed and the remaining frequencies were tested applying a χ ² test. Further, the ploidy and growth form data were fitted to a generalized linear model using R for Mac version 2.14.0 (R Development Core Team 2011), where the observed counts of the different ploidy and growth form combinations were the response variables and ploidy and growth form were the predictor variables. Best model fit was achieved by specifying the model with a negative binomial error distribution. In order to characterize the habitats and ecological amplitude of S. oppositifolia, point frame data of the vegetation around or in the vicinity of each sample, and the associated environmental variables, were analysed with ordination techniques applying the R package Vegan 2.0–3 (Oksanen et al. 2012). Three basic ordination methods were applied: non-metric multidimensional scaling (NMDS), principal component analysis and detrended correspondence analysis (DCA). Spearman's rank correlation coefficients were determined for all environmental variables that had a significant influence in the preferred ordination analysis. In addition, these variables were tested for significant differences between the two main ploidy levels and the three growth forms. For the differences according to ploidy levels, a Wilcoxon signed-rank test was used, and for differences according to growth forms, a Kruskal–Wallis one-way analysis of variance was applied. The series-packing module in PAST (Hammer et al. 2001) was used to fit Gaussian response models to the abundances of different ploidy levels of S. oppositifolia along each of the environmental gradients showing significant differences between ploidy level and/or growth form. In these analyses, it was assumed that all point frame hits of S. oppositifolia within a vegetation plot had the same ploidy level and growth form as the sampled plant with which the vegetation plot was associated.

Ploidy level could be successfully determined in 193 of the 195 collected samples. Our results reconfirmed the presence of three ploidy levels in S. oppositifolia in Svalbard, with a majority of diploids and tetraploids (see Müller et al. 2012; Fig. 1): 106 (55%) were diploids, eight (4%) were triploids and 79 (41%) were tetraploids. The geographical distribution of ploidy levels showed that both diploids and tetraploids are widespread in Svalbard, and no clear overall distribution pattern was found (Fig. 1). Diploids and tetraploids co-occurred within 11 sampling sites, while eight sampling sites were monotypic (Table 1). The different growth forms also seemed to be widespread and often intermixed, but separate analyses of the locations where both ridge and slope habitats were sampled showed that ploidy level (χ ²=16.83, df=2; P<0.0002) and growth form (χ ²=14.494, df=2; P<0.001) were strongly related to habitat. Tetraploids and the prostrate growth form were clearly more frequent in slope habitats, while the diploids and the cushion growth form were clearly more frequent on ridges (Table 2).

While all three growth forms were observed among diploids, no cushion forms were observed in triploids, and only two of the 79 tetraploids, were reported as cushion-shaped (Table 3). The frequency distributions of growth forms and ploidy levels were significantly different in all three performed tests. For the full contingency table with all growth forms and ploidy levels, Cramér's V was 0.508 with χ ²=99.596 (df=4; P<0.0001). The deviations from independence of rows and columns in the two dimensional contingency table are shown in the Cohen-Friendly association plot (Fig. 2). After removing the triploid data, the χ ²=94.357 (df=2; P<0.0001). The results from the generalized linear model confirmed the two former tests, and showed a significant difference in the count of diploids and tetraploids forming cushions. Further, the model showed a significantly higher count in prostrate tetraploids (Table 4).

All ordination methods showed similar results, but we gave preference to the NMDS. This method has the advantage over other ordination techniques in that it makes only a few assumptions about the nature of the data, for example, species responses are not restricted to linear relationships (Oksanen 2011). Some of the measured environmental variables had no significant influence (soil moisture) or very low influence (R² below 0.1; bare ground and litter) on the vegetation structure. Based on the NMDS, the most relevant environmental variables structuring the vegetation were soil temperature, lichen cover, LOI, pH and bryophyte cover (Supplementary Table S1). All environmental variables were further tested for differences between plots with different ploidy levels and different growth forms. Lichen cover, bare ground, soil pH, soil temperature and vascular plant cover were significantly different between plots with different ploidy level and different growth form. Cover of litter was only significantly different between plots with different growth form (Supplementary Table S2). The correlation between environmental variables was high for several variables (Supplementary Table S3). Soil pH was closely negatively correlated to lichen cover and LOI (Rs=−0.59, P<0.001; Rs=−0.50, P<0.001). Bare ground and vascular plant cover were also strongly negatively correlated (Rs=−0.55, P<0.001).

A NMDS plot was drawn including all environmental variables (Fig. 3). Two sampling sites in Wijdefjorden (Location 5; Table 1) had a strong deviation in species composition compared to the other sampling sites and were excluded from the final multivariate analysis in order to achieve a better resolution of the remaining sampling sites. The NMDS plot clearly showed an ecological difference between ploidy levels. Vegetation plots associated with tetraploids were clustered together, while the diploids appeared more scattered. The micro-habitat associated with tetraploids was clearly characterized by higher soil pH, higher vascular plant cover and higher soil temperatures. The few scattered tetraploids situated at the lower end of NMDS axis 1, were samples from Florabukta. Some of these samples were collected close to a bird cliff, and were the only vegetation plots sampled in manured polar desert.

Gaussian response models of diploids versus tetraploids reflected the ecological differences found in other analyses. The environmental variables showing differences in optimum values for diploids versus tetraploids were soil pH (6.53 vs. 7.65; Fig. 4), soil temperature (8.8°C vs. 12.0°C), bare ground (21.2% vs. 4.0%) and vascular plant cover (29.4% vs. 46.3%). Therefore, diploids seemed to prefer lower pH, less vegetation cover and lower temperatures compared to the tetraploids. Diploids showed also higher tolerance for all tested environmental variables (soil pH=±1.14 vs. ±0.24; Fig. 4; soil temperature=±3.1°C vs. ±2.2°C; bare ground; ±28.1% vs. ±9.2%; vascular plant cover=±15.9% vs. ±14.4%). Doing the same analyses sorting by growth form (cushions versus prostrate) gave mainly the same values and results. Cushions preferred more bare ground, less vegetation cover and lower temperatures and had overall higher tolerance levels. The exception was for soil pH, where only minor differences were found between growth forms (7.34±0.73 vs. 7.48±0.62).