Study Area

The study was conducted in the north-western part of the Sierra Nevada Mountains, south-eastern Spain. The study area consisted of five sites on a single slope up to the Veleta peak (3396 m; 37°03′ N/03°21′ W), at elevations of 2315, 2520, 2720, 3010 and 3240 m. The distance between the two neighbouring sites was c. 1.5–2 km. The two highest elevation sites are characterized by open grasslands and plants with basal rosettes such as Festuca clementei, Erigeron frigidus or Jasione amethystina, whereas the three lower sites are over moderately developed soils characterized by species such as Festuca indigesta, Leontodon boryi or Sideritis glacialis (Valle 2003). In Pradollano (2507 m a.s.l.) close to our 2520 m site, mean annual rainfall is 690 mm, and mean annual temperature is 3.9 °C (Rivas-Martinez & Rivas-Saenz 1996–2009). The study area is, therefore, characterized by a Mediterranean climate including a relatively hot and dry summer spanning the whole growing season. Correlations between mean annual temperature or precipitation and elevation in the northern side of the Sierra Nevada range revealed a temperature decrease of 0.61 °C every 100 m and an increase in precipitation of 33.3 mm across the same elevation interval (Delgado Calvo-Flores et al. 1988), thereby resulting in a crossed climatic gradient along the elevation gradient (Fig. 1a).

All five sites were situated in siliceous gravel habitats with gentle slopes facing north-west and containing the cushion-forming species Arenaria tetraquetra ssp. amabilis (Bory) H. Lindb. fil. (Caryophyllaceae), a species endemic to the Sierra Nevada Mountains (see Fig. S1 in Supporting Information). Arenaria tetraquetra cushions appeared haphazardly dispersed over the study sites and showed no obvious microsite preferences. Large open areas surrounded each selected cushion. The relative cover of A. tetraquetra was determined by five 30-m linear transects: the proportion of intercepted transect length of A. tetraquetra cushions gave a measurement of cover, which ranged between < 1 and 4%. The relative cover of all other vascular plants in open areas was visually estimated in each sampled open plot, and varied between sites from c. 3% to 14% on average, and was generally higher at low elevation than at high elevation. Our five sites encompassed the whole elevation range of A. tetraquetra, which was estimated to grow between 2400 and 3300 m a.s.l. (Díaz de la Guardia 2007). Both cushions and open areas were colonized by small annuals and (predominantly) by perennial herbs and grasses with a high frequency of endemism to the Sierra Nevada Mountains (> 40%; C. Schöb, pers. obs.).

Diversity Sampling

We sampled 30 cushions and 30 paired neighbouring open areas at 2315, 2520 and 3010 m, whereas at 2720 and 3240 m, the number of pairs was 85 and 100, respectively, giving 550 samples in total. At each site, all samples were taken within an area of c. 500 m². Hereafter, we will refer to each sampled cushion and its paired open area as plots. Samples of open areas were equal in size and shape to that of cushions and were randomly placed at c. 50 cm distance to the respective cushion. Plot size was determined by measuring the largest cushion diameter and its perpendicular length, and its area was calculated as π×((diameter₁+diameter₂)/4)². Therefore, plot size varied between samples due to variability in cushion size but was always the same between a cushion and its paired open plot. Average plot size was 174 ± 5 cm² (mean ± 1 SE). In July 2010, at the peak of the growing season, the number of species and the number of individuals (or shoots for clonal species) of all vascular plant species were recorded within the whole area of both cushions and open plots. We quantified plant communities in each plot using both the number of species and the number of individuals: the effects of nurse plants on community assembly have been shown to influence the composition of species but also the number of co-occurring individuals of the same and different species (Schöb, Butterfield & Pugnaire 2012). The number of species ranged between 0 and 9 per plot, and the number of individuals ranged from 0 to 85 per plot. The total number of species recorded over all sites was 41 (see Table S1 in Supporting Information).

Microenvironmental Data

Between 20 and 21 July 2010, we sampled 15 cm³ of soil from the uppermost 10 cm beneath cushions and in open areas in six pairs of cushions and open plots per site (10 pairs at 2720 and 3240 m) and determined soil water (SWC) and soil organic matter (SOM) content. Soil samples were stored in sealed containers and fresh mass determined immediately after sampling. Gravimetric SWC (g g_soil⁻¹) was measured by mass loss after drying at 105 °C for 48 h. SOM (g g_soil⁻¹) was determined in the same soil samples by mass loss after ignition at 400 °C for 20 h.

At three sites (2315, 2720 and 3240 m), we recorded air temperature with iButtons (Maxim Integrated Products, Sunnyvale, CA, USA). Three temperature sensors per site were randomly placed in open areas 5 cm above-ground and protected from direct solar radiation. Temperature was recorded at 1-h intervals. For subsequent analyses, we calculated mean, maximum and minimum daily temperatures during August 2010.

Plant Functional Traits

We assessed variation in 10 functional traits of A. tetraquetra cushions that were expected to respond to environmental conditions along the elevation gradient, and which potentially affect interactions with other plants. We used five traits as proxies of physiological status: (i) leaf dry matter content, LDMC (g kg⁻¹), measured as the ratio between leaf dry mass and fully rehydrated fresh mass; (ii) specific leaf area, SLA (m² kg⁻¹), measured as the ratio between fresh leaf area and leaf dry mass; (iii) relative water content, relative water content (RWC) (%), determined as the ratio between plant water content in the field and water content after rehydration at full turgor; (iv) effective quantum yield (∆F/F_m′), the yield of photochemical energy conversion at photosystem II (PSII) reaction centres; (v) apparent electron transport rate, ETR (μmol m⁻² s⁻¹), the efficiency of electron transportation at PSII. All these traits are considered indicative of the response of plant physiological status to local environmental conditions (Mooney, Hillier & Billings 1965; Noy-Meir 1973; Körner 2003; Reverter et al. 2010).

We also measured five functional traits for use as proxies of plant morphology: (i) lateral spread, measured as canopy diameter (cm) and used as a proxy for plant size; (ii) cushion thickness, determined by penetrating the cushion with a metallic rod and measuring the mean distance (cm) between the tip of the outer leaves and the soil underneath the plant at five random locations within the cushion; (iii) branch density (branches cm⁻²), calculated from the number of terminal branches counted in one randomly placed 2 cm × 2 cm quadrat per cushion; (iv) terminal branch length, measured as the distance (mm) between the bottom of the terminal leaf rosette and the terminal branching (see also Fig. S1d); and (v) leaf density (leaves cm⁻²), calculated as the average number of leaves per terminal branch of ten randomly selected branches multiplied by the number of branches cm⁻². The area of individual leaves of A. tetraquetra was generally very small (2.6 ± 0.2 mm²) and showed low variability, both within and between sites. Due to the low responsiveness of this trait in our study system, it was not considered in analyses. High values of branch and leaf density and low values of terminal branch length indicate high cushion compactness. Compactness, lateral spread and thickness together indicate cushion size and biomass. These functional traits were, on the one hand, considered dependent on the physiological status of the plant (Milla, Giménez-Benavides & Montserrat-Marti 2008), but on the other hand, they were considered to influence microenvironmental conditions within the cushion (Crutsinger, Strauss & Rudgers 2010; Michalet et al. 2011). All these traits were measured on 10 cushions per site between 16 and 18 July 2010.

LDMC and SLA were determined in leaves of five cushions per site collected between 9 and 10 August 2010. Following Cornelissen et al. (2003), we selected one small branch per cushion containing mature and healthy leaves and fully rehydrated them before determining leaf area and saturated mass of two leaves. Dry mass was measured after 72 h in the oven at 70 °C. LDMC and SLA are generally associated with photosynthetic capacity, relative growth rate and leaf longevity (Cornelissen et al. 2003), but are regulated by construction costs related to plant water and carbon economies, respectively. LDMC generally decreases with greater water availability and lower likelihood of physical damage (Cornelissen et al. 2003) and often shows no response to temperature variation (Albert et al. 2010). By contrast, SLA increases in response to increasing temperatures and decreasing light availability across a broad range of species (Poorter et al. 2009) and is positively correlated with soil nitrogen availability (Ordoñez et al. 2009) and negatively correlated with leaf longevity and resource-use efficiency (Wright et al. 2004). Thus, whilst LDMC and SLA are associated with similar biological processes, they may vary independently in response to different environmental drivers.

Relative water content measurements were performed on branches from eight cushions per site between 20 and 21 July at midday (11.00–13.00, solar time) under cloudless conditions and with cushion surface air temperatures of 21 ± 0.2 °C (mean ± 1 SE). Time of sampling, temperature and their interaction term did not affect measurements. Fresh branch mass was determined immediately after sampling in the field, saturated mass was measured after 30 h of rehydration in the dark, and dry mass was measured after 72 h at 70 °C. Accordingly, plant water content in the field was calculated as the difference between field fresh mass and dry mass, and water content at full turgor was calculated as the difference between saturated mass and dry mass. RWC was used as an indicator of plant water status at the time of sampling.

Effective quantum yield (∆F/F_m′) and ETR were determined in 10 randomly selected cushion plants per site measured with a saturation-pulse fluorometer (Mini-PAM Photosynthesis Yield Analyzer; Walz Mess- und Regeltechnik GmbH, Effeltrich, Germany) under ambient radiation and steady-state conditions. Measurements were carried out in the centre of the cushion on areas with no apparent leaf damage and with similar appearance amongst all cushions. F_m′ is the maximum fluorescence of the leaf adapted to ambient irradiation induced by application of a saturating light pulse (12 000 μmol photons m⁻² s⁻¹) closing all PSII reaction centres. ∆F = F_m′−F, where F is the steady-state fluorescence (Schreiber & Bilger 1993). ETR was calculated after Krall & Edwards (1992) combining the effective quantum yield and the corresponding photosynthetically active radiation (PAR):

ETR=(ΔF/Fm')×PAR×0.5×0.84(eqn 1)

where 0.5 takes into account that about 50% of all absorbed photons reach PSII, and 0.84 is considered the most common leaf absorbance coefficient for C3 plants (Björkman & Demmig 1987). Effective quantum yield and ETR measurements were performed between 20 to 21 July at midday (11.00–13.00, solar time) under cloudless conditions and at air temperatures close to the cushion surface of 21 ± 0.2 °C (mean ± 1 SE). Time of sampling, temperature and its interaction term did not affect the measurements. Both fluorescence measures are indicators of the efficiency of photosynthesis at the time of sampling either taking differences in incoming PAR into account (i.e. ETR) or not (i.e. effective quantum yield), and can be considered rough measurements of leaf stress.

Data Analysis

We quantified the absolute and relative difference in SWC and SOM between cushions and open areas, where the relative difference was calculated by dividing absolute differences by the corresponding values in open plots. The difference in the number of species and individuals between cushions and open areas was calculated using the relative interaction index (RII; Armas, Ordiales & Pugnaire 2004). RII was calculated for each paired cushion and open plot as

RII=(Nc−No)/(Nc+No)(eqn 2)

where N_c is the number of species or individuals in cushions, and N_o is the corresponding value in the open plot.

For all statistical analyses, we applied a model selection approach using the Akaike Information Criterion (AIC), where lower AIC values indicate a better fit of the corresponding model based on the concept of parsimony, that is, by balancing complexity and power of a model (Johnson & Omland 2004). In addition, we calculated Akaike weights giving the likelihood of each model to be the best model in comparison with all the competing models tested. As some of our models, in particular those including functional traits, include a high number of explanatory variables, we preferred a model selection approach over traditional null hypothesis testing. It allowed us to compare several competing models and to select the best model that included only those variables, which substantially improve the model fit. Furthermore, it allowed us to rank models that included different combinations of variables, thereby measuring the relative support for each competing model.

The relationships between either SOM or SWC and the presence/absence of cushions and elevation were tested with generalized linear models with a Poisson error structure and a log link function. The relationships between either the number of species or the number of individuals and the presence/absence of cushions were tested with generalized linear models with a Poisson error structure, a log link function and elevation included as a covariate. Comparisons of the number of species and individuals along the elevation gradient were omitted due to unequal plot sizes at different elevations. Linear regression was used to test whether absolute or relative differences in SOM, SWC, number of species (RII_species) and number of individuals (RII_individuals) between cushions and open areas varied along the elevation gradient.

For the five functional traits used as proxies for plant morphology, we applied a principal components analysis (PCA) with Varimax rotation (Quinn & Keough 2002) to reduce the number and multicollinearity of variables. We used a correlation matrix (i.e. zero-centred variables with unit variance) because variables were in different units and had different variances. The principal components (PCs) were, then, used as predictor variables in linear models testing the relationship between cushion morphology and differences in SOM or SWC between cushion and open plots. We included SOM or SWC in open plots as covariates in the corresponding models. These covariates accounted for those changes in the differences in soil properties between the two microhabitats with elevation that are due to changes in SOM or SWC with elevation in the open microhabitat, and are therefore not related to the effect of cushions. The models consisting of all possible combinations of PCs and the covariate were then compared with the control model (a model only including the covariate). If any model including a PC and the covariate performed better than the control model, there was support for a relationship between cushion morphology and the differences in either SOM or SWC between cushions and open plots. This relationship was used as indication for a role of cushion morphology in the potential effects of cushions on microenvironmental conditions. With similar models, we tested for the relationship between cushion morphology and RII_species or RII_individuals where we included the number of species or individuals in open plots as a covariate. This relationship was used as an indication for the role of cushion morphology in potential facilitation effects of cushions on species richness and abundance.

Because functional traits were not directly matched to specific values of SOM, SWC, number of species or number of individuals within sites (i.e. these variables were sampled in different cushions within the same sites), we only studied their relationship across sites using permutation tests that randomly attributed to each measurement of a morphological trait a measurement of soil properties or the mean of three measurements of the diversity sampling. Random sampling occurred without replacement and was restricted within sites. Each of these data sets was equal to the minimal size of the observed data set (i.e. n = 38 and n = 50 for the relationship between functional traits and soil properties and between functional traits and RII_species and RII_individuals, respectively). With this procedure, we kept differences in functional traits, soil properties, number of species and number of individuals amongst sites but randomized them within sites. We preferred this procedure over calculating mean values per site for each variable in that it enabled us to include more complex models testing for relationships between those variables amongst sites. We performed 1000 randomizations of the data sets and calculated AIC values, correlation coefficients between variables and adjusted R² values for each candidate model for each randomization. The best model was, then, selected according to the lowest mean AIC value (AIC_mean). In addition, we calculated Akaike weights for all candidate models based on AIC_mean, the mean correlation coefficient (r) between each PC and the response variable as a measure of effect size of each PC on the response variable and mean adjusted R² (Radj 2) values for each model as a measure of the goodness-of-fit of the corresponding model. Finally, the relationship between morphological and physiological traits of cushions across sites was revealed by Pearson correlations based on mean values per site (n = 5).

All calculations and statistical analyses were performed with the r software version 2.14.0 using the packages base, graphics, grDevices, lme4, psych, stats and vegan (R Development Core Team 2011).

Microenvironment

In open areas, there was an increase in SWC with elevation, whereas SOM and temperature decreased with elevation. SWC in the lowest site was 1.4% of that in the highest site, whereas SOM decreased by 67% from the lowest to the highest site (Fig. 1b). T_mean was lower in the highest site (13.5 ± 0.3 °C vs. 19.9 ± 0.4 °C), in parallel with T_min (7.2 ± 0.1 °C at 3240 m and 11.7 ± 0.4 °C at 2315 m) and T_max (24.1 ± 0.7 °C at 3240 m and 33.1 ± 2.1 at 2315 m).

Both SOM and SWC changed not only with elevation but also with the presence of cushions, showing higher values beneath cushions than in open areas (Fig. 2a,b). The best model explaining SOM included the two main effects, ‘elevation’ and ‘cushion’ (Radj 2 = 0.31, Table S2), whereas SWC was best explained with a full model including ‘elevation’, ‘cushion’ and their interaction term (Radj 2 = 0.59, Table S2). The absolute difference in SWC between cushions and open plots increased with elevation (Radj 2 = 0.21), but this pattern was not found for SOM (Fig. 2a,b; statistical results in Table S3a). However, the relative differences between cushions and open areas increased with elevation for SOM (Radj 2 = 0.10) but not for SWC (statistical results in Table S3b).

Diversity

Cushions harboured a higher number of species and individuals per plot than open areas at all sites (with the exception of number of individuals at 2520 m; Fig. 2c,d; Table S4). RII_species and RII_individuals were lowest at intermediate elevation (2720 m), increased towards higher elevations and slightly increased towards lower elevations (Fig. 2e,f). Accordingly, the best models fitting RII_species (Radj 2 = 0.07) and RII_individuals (Radj 2 = 0.15) consisted of the main effect ‘elevation²’ (Table S5) indicating a nonlinear relationship.

Physiological Traits and Cushion Morphology

There was a negative relationship between elevation and LDMC (r = −0.40) and SLA (r = −0.22) and a positive correlation between elevation and RWC (r = 0.23), effective quantum yield (r = 0.20) and ETR (r = 0.72; Fig. S2). LDMC and SLA showed negative correlations with all morphological traits except terminal branch length, whereas RWC, effective quantum yield and ETR showed positive correlations for all traits except terminal branch length (Fig. 3 and Table S6). The strongest relationships were found between morphological traits and ETR (r > 0.75 for all morphological traits; Table S6).

Cushion Morphology and Environment/Diversity

All functional traits used as proxies for cushion morphology varied with elevation. Cushions were larger (r = 0.64), thicker (r = 0.41), had higher branch and leaf density (r = 0.87 and r = 0.88, respectively) and shorter terminal branch length (r = −0.57) at high than at low elevations (Fig. 3 and Fig. S2). The first three axes of the PCA with the five morphological traits explained 58%, 24% and 10% of the variance, respectively. The first axis (PC1) represents cushion compactness and size, with strong positive correlations for branch density, leaf density and lateral spread (Fig. 3 and Table S7). The second axis (PC2) represents cushion thickness and branch length with strong positive correlations for thickness and terminal branch length. Positive values on the third axis (PC3) represent small and rather compact cushions with negative correlations for lateral spread and positive correlations for leaf density.

Differences in the number of species and individuals between cushions and open areas were best explained by PC1 with minor contributions of PC2 and PC3 (Tables S8 and S9). Both RII_species and RII_individuals showed positive correlations with PC1 (r = 0.25 and r = 0.36, respectively). PC1 was also the best predictor for the absolute differences in SOM between cushions and open areas, with minor contributions of PC2 and PC3 (Table S10), whereas the model including PC1 and PC3 best predicted absolute differences in SWC between cushions and open areas (Table S11). PC1 was negatively correlated with differences in SOM between cushions and open areas (r = −0.12), whereas all principal components were positively correlated with differences in SWC between cushions and open areas, with a particularly high effect size for PC1 (r = 0.46).

Cushion Traits and Facilitation

The cushion life-form is most frequent in nutrient poor and cool environments, particularly on bare soils or in open vegetation in windswept habitats (Körner 2003). In these environments, the compact and dense canopy of cushions traps and accumulates leaf litter (Körner 1993), mitigates extreme temperatures inside the cushion and increases soil moisture beneath it (Cavieres et al. 2006). This creates a favourable microenvironment for decomposers (Schinner 1982a), which probably results in the cushion acting as a fertility island (Körner 2003). However, it is highly likely that all these components of microhabitat amelioration by cushions are highly dependent on their morphology. More compact and larger cushions probably provide a better wind shield than loose and small cushions, thereby retaining leaf litter, ameliorating low temperatures and increasing moisture storage. In line with this, the effects of cushions on SOM and SWC were highest at high elevation where cushions were compact and large, whereas at low elevation, where cushions were loose and small, these effects were smaller. Consequently, cushions may need a certain range of trait values to efficiently improve microhabitat conditions and act as nurses.

The positive link between physiological and morphological traits of A. tetraquetra points towards a dependence of the cushion morphology on its vigour. A better physiological status (e.g. higher ETR, effective quantum yield and lower SLA) and an improved water balance (e.g. higher RWC and lower LDMC) went along with compact and large cushions. Arenaria tetraquetra developed its optimal physiological status at high elevations. At low elevations – with low water availability and high temperatures – A. tetraquetra performed poorly, as indicated by low plant water status and low photosynthetic efficiency. Indeed, Reverter et al. (2010) showed that the permanent wilting point for most plants growing in the Sierra Nevada Mountains at 2300 m occurred when gravimetric soil water content dropped below 3%; SWC at our low site, both in open areas and beneath cushions, was below this threshold. Therefore, at low elevations, A. tetraquetra was most probably too stressed to develop large cushions with a compact and dense canopy.

The benefit of the microhabitat amelioration (sensu Callaway 2007), or the creation of biogenic microhabitats (sensu Badano & Marquet 2009) by cushions, is not necessarily restricted to other species but also probably benefits the cushion species itself. Consequently, we assume that strong facilitation occurred under conditions where niche construction (sensu Kylafis & Loreau 2011) by A. tetraquetra was adequate, as occurred at high elevation. At low elevations, however, A. tetraquetra may not have been able to adequately develop its own niche because the most limiting niche factor (i.e. water availability) could not be considerably improved. Overall, these results highlight the importance of niche construction by cushions for their facilitation effect.

Facilitation Along Opposing Stress Gradients

Changes in facilitation intensity – that is, the difference in species richness and the number of individuals inside and outside cushions – along the elevation gradient were likely to be a reflection of the opposing stress factors in our study system that affected species richness and abundance both in and out of cushions. On the one hand, our results suggest that the number of species and individuals associated with A. tetraquetra changed with elevation through the effects of the opposing stress factors on cushion morphology – the latter affecting the influence of the nurse on microenvironmental conditions. On the other hand, the opposing stress gradients directly affected species richness and abundance in open areas, that is, a reduced richness and abundance with increasing environmental severity towards both ends of the elevation gradient.

Facilitation intensity was high at the high end of the elevation gradient but not at the low end. This difference in facilitation intensity at the two stressful ends of the gradient could have been the result of differences in the capacity of the nurse to mitigate the different environmental factors affecting the interaction balance (Michalet 2007; Maestre et al. 2009). Environmental severity at high elevation was characterized by non-resource-related stress (i.e. low temperature), whereas at low elevation, it was caused by resource limitation (i.e. low water availability). However, low temperature at high elevation may indirectly affect nutrient availability via reduced plant productivity and SOM (Körner 2003), especially because SOM is the primary source of soil nutrients in our study system (Sánchez-Marañón et al. 2002). A case study performed at our field site showed that SOM is positively correlated with total N (Sánchez-Marañón et al. 2002), and similar relationships between SOM and total N have also been shown in other alpine (Rehder 1970; Körner 2003) and water-limited (dry to arid) ecosystems (Whitford, Reynolds & Cunningham 1987). Furthermore, low temperatures and short growing seasons in high elevation sites are associated with low microbial activity, which may result in low decomposition rates of SOM, which would further reduce soil nutrient availability at high elevations (Schinner 1982b; Seastedt, Walker & Bryant 2001). Consequently, at both ends of the elevation gradient, the limiting factors for plant growth may have been resources: nutrients at high elevation and water at low elevation. Nonetheless, at high elevations, soils beneath A. tetraquetra cushions showed higher levels of resources (i.e. SWC and SOM) than those in open areas, indicating that cushions likely buffered this resource-related stress at high elevation. This is in line with two previous reports on cushion plants in alpine environments that revealed nutrient enrichment at high elevation sites (Yang et al. 2010; Anthelme et al. 2012) and provision of moisture in arid environments (Cavieres et al. 2006) as drivers of increased facilitation. Consequently, these results indicate that A. tetraquetra was able to alleviate both limiting stressors provided that it could develop the appropriate combination of functional trait values needed to improve microenvironmental conditions and act as nurse.