Explore the vast range of titles available.

Summary The stress‐gradient hypothesis (SGH) predicts that the balance of plant–plant interactions shifts along abiotic environmental gradients, with facilitation becoming more frequent under stressful conditions. However, recent studies have challenged this perspective, reporting that positive interactions are, in some cases, more common at the intermediate level of environmental severity gradients. Here, we test whether and how neighbour effects by Silene acaulis cushions vary along a 700 m wide altitudinal transect, in relation to cushion morphological traits and environmental severity. Field measurements along the gradient, within and outside cushions, included (i) species richness and cover of coexisting vascular plants; (ii) cushion morphology; (iii) above‐ and below‐ground microclimate; and (iv) soil quality. We used the relative interaction index to decouple neighbour trait effects and environmental severity effects on plant diversity at different elevations. The ability of the cushion plant to facilitate heterospecifics shifts considerably along the elevation gradient, being greatest at the intermediate level. On the other hand, Silene morphological traits steadily change along the gradient, from lax, soft and flat‐shaped cushion habits at low elevation to tightly knit and dome‐shaped habits at high elevation. Cushion morphological changes are associated with mitigating effects on microclimate, indicating that cushions effectively act as a heat‐trap at medium and high elevations, while at low elevations the soft and flat cushions avoid excessive heat accumulation by tight coupling with the surrounding atmosphere. At the upper end of the gradient, cushion cespitose–pulvinate compactness and high stem density appear to be critical traits in modulating the net effect of plant–plant interaction, since the space available for hosting other vascular species is considerably reduced. In conclusion, this work provides a mechanistic link between plant morphological traits, associated biogenic microclimate changes and variation in net plant–plant interactions along the explored severity gradient. Our findings support an alternative conceptual model to SGH, with plant facilitation collapsing at the upper extreme of the abiotic stress gradient. Lay Summary

The stress-gradient hypothesis (SGH) predicts that the community-wide prevalence of positive interactions, relative to negative interactions, is greater under more severe environmental conditions. Because the frequency of positive and negative interactions within a community is the aggregate of multiple pair-wise interactions, one approach to testing the SGH is to examine how pair-wise interactions vary along severity gradients. While the SGH suggests that the net outcome of an interaction should monotonically become more positive with increasing environmental severity, recent studies have suggested that the severity-interaction relationship (SIR) may rather be unimodal. We tested which of the proposed shapes of the SIR best fits the variation in the interaction between two species along two types of severity gradients on sub-Antarctic Marion Island. This was done by comparing the performance of the grass Agrostis magellanica in the presence and absence of the cushion plant Azorella selago, along both species' entire altitudinal ranges (transects spanning 4-8 km), and along a shorter (transect = 0.4 km) wind exposure gradient. Along the altitudinal transects the relative intensity, but not the absolute intensity or the importance, of the Azorella selago-Agrostis magellanica interaction increased with altitude, consistently forming a plateau-shaped SIR with a positive asymptote. Thus, while the performance of Agrostis magellanica was negatively affected by Azorella selago at low altitudes, the grass benefited from growing on the cushion plant under greater environmental severity. Along the wind exposure gradient the intensity of the interaction also became more positive with increasing environmental severity for most performance measures. This suggests that the switch from a net negative to a net positive interaction can occur across both short and long distances. Therefore, this study provides strong evidence for a plateau-shaped SIR, and confirms that the SIR is unimodal along the particular non-resource severity gradients of this study.