The shift from plant–plant facilitation to competition under severe water deficit is spatially explicit

Abstract The stress‐gradient hypothesis predicts a higher frequency of facilitative interactions as resource limitation increases. Under severe resource limitation, it has been suggested that facilitation may revert to competition, and identifying the presence as well as determining the magnitude of this shift is important for predicting the effect of climate change on biodiversity and plant community dynamics. In this study, we perform a meta‐analysis to compare temporal differences of species diversity and productivity under a nurse plant (Retama sphaerocarpa) with varying annual rainfall quantity to test the effect of water limitation on facilitation. Furthermore, we assess spatial differences in the herbaceous community under nurse plants in situ during a year with below‐average rainfall. We found evidence that severe rainfall deficit reduced species diversity and plant productivity under nurse plants relative to open areas. Our results indicate that the switch from facilitation to competition in response to rainfall quantity is nonlinear. The magnitude of this switch depended on the aspect around the nurse plant. Hotter south aspects under nurse plants resulted in negative effects on beneficiary species, while the north aspect still showed facilitation. Combined, these results emphasize the importance of spatial heterogeneity under nurse plants for mediating species loss under reduced precipitation, as predicted by future climate change scenarios. However, the decreased water availability expected under climate change will likely reduce overall facilitation and limit the role of nurse plants as refugia, amplifying biodiversity loss.

Indeed, facilitation is often observed in systems that are experiencing growth-or fitness-limiting climatic conditions, such as hot and dry semiarid ecosystems or savannas (Butterfield et al., 2016;Dohn et al., 2013;Pugnaire et al., 2011). Therefore, facilitative interactions may promote species diversity by maintaining species in conditions outside their fundamental niche (Bruno, Stachowicz, & Bertness, 2003). This greater species diversity is important for ecosystem functions such as productivity and ecosystem stability (Hector et al., 1999;Isbell et al., 2015;Tilman, Reich, & Knops, 2006). This stabilizing effect may be especially important for semiarid ecosystems that are already experiencing important climatic changes, which are pushing the environmental conditions to novel extremes (Giorgi & Lionello, 2008;Lloret, Siscart, & Dalmases, 2004). As an example, the yearly rainfall of the Mediterranean Basin is expected to decrease as much as 20% within the next 50 years, while temperatures are likely to increase (IPCC 2014). In conjunction with these gradual changes, extreme events such as long heat waves and heavy precipitation events are predicted (Lloret et al., 2004). Combined, these climatic shifts have potentially important consequences for the currently prevailing facilitative interactions found in semiarid systems and the biodiversity and ecosystem functions that depend on those positive plant-plant interactions (Brooker et al., 2008;Mueller et al., 2005).
In this study, we compare the relative effect of nurse plants on species diversity (richness) and community productivity (plant biomass) of vascular plants in a range of years varying from low to high rainfall (Table 1) to assess nurse plant effects on herbaceous communities across a range of water availability. We also tested spatial differences in diversity, productivity, abundance, and composition of the herbaceous plant communities in the four cardinal directions around nurse plants during a year with severely low rainfall. This detailed spatial assessment was carried out under a species (Retama sphaerocarpa) that has been well documented to facilitate diversity at this site under average climatic conditions Pugnaire & Lázaro, 2000;Pugnaire et al., 1996). Our hypothesis was that severe water limitation promoted negative plant-plant interactions but the magnitude of that effect depended on variation in microsite conditions under the nurse species.

| Study site
Fieldwork was conducted in a Retama sphaerocarpa shrubland located in Rambla del Saltador, Almería Province, SE Spain (37°08′N, 2°22′W) at 630 m elevation. Climatic conditions are semiarid with a dry summer season between June and September. Mean annual rainfall is ~250 mm and annual mean temperature is 16°C (Pugnaire & Lázaro, 2000). The soil at the valley bottom is a sandy loam of alluvial origin with poor water-holding capacity and low soil organic matter and nutrient concentrations (Pugnaire et al., 1996) (Pugnaire et al., 1996). This generally facilitative effect of the shrub is attributed to higher water availability (Prieto et al., 2010;Pugnaire, Armas, & Valladares, 2004) as well as higher soil organic matter and nutrient concentration (Pugnaire et al., 1996

| Sampling design
In

| Abiotic conditions
We measured photosynthetically active radiation (PAR), soil temperature, and soil depth within each part of the sampling grid. PAR was measured 10 cm above ground with a LS-C Mini Quantum Sensor (Heinz Walz GmbH, Germany). Soil temperature was measured at 2 cm below the soil surface with handheld thermometers. Soil depth was measured with a penetrometer that was inserted into the soil until it could not be introduced further without reasonable effort. Soil depth was measured as the length of the penetrometer that was inserted into the soil. Plots in open locations were always measured immediately after the associated plot beneath Retama in the same aspect to reduce temporal differences in temperature and light within each pair of Retama and open plots (see Table S1 in Supporting Information).

Relative interaction intensity (RII) for both species richness and
productivity shifted from positive (i.e., facilitation) to negative (i.e., competition) as total rainfall decreased. However, the significant categorical rainfall variable-fit after the continuous term (Table   S2)-indicates that RII did not respond linearly to decreasing rainfall.
Therefore, below a distinct rainfall quantity RII became significantly negative (Figure 2a

| Response of facilitation to rainfall
In support of the hypothesis that facilitation shifts to competition under severe resource limitation (Michalet et al., 2014(Michalet et al., , 2015, RII for species richness and community productivity shifted to negative between 70 and 120 mm, while nearly universal positive effects were found above 120 mm. Our results directly support the conceptual model proposed by Tielbörger and Kadmon (2000) whereby under nurse plants water limitation occurs earlier than in open locations and beneficial effects such as improved nutrients or shading are overshadowed by competition for water. We propose two hypotheses to explain this threshold effect: (1) Competition for water under Retama is higher than in open locations during severely low rainfall years (Butterfield et al., 2016), or (2) species adapted to live under Retama have less tolerance to severe drought (Tielbörger & Kadmon, 2000). However, additional biotic and abiotic variables interact with rainfall to alter this threshold (Butterfield et al., 2010). For example, larger Retama plants increase niche space and microsite variability thereby often supporting higher diversity and productivity than smaller Retama (Pugnaire et al., 1996;Schöb et al., 2013) and delaying the effects of water limitation through increased shading. Differences among sampling designs and other climatic variables from the multiple years may have affected the results, but these artifacts were hopefully controlled for with the use of RII. Regardless, hotter and drier climatic conditions are expected in many Mediterranean ecosystems, which will continue to promote decreases in species richness and productivity.

| Spatial differences in diversity, productivity, and composition
Overall plots under Retama failed to facilitate more diverse and productive understory communities during a severely low rainfall year, and the near absence of plants in the south and the east aspects indicates that Retama had a negative effect on community diversity and productivity (Michalet et al., 2015), further supporting the argument that facilitation gives way to competition under severe water limitation (Michalet et al., 2014;Tielbörger & Kadmon, 2000). In contrast to the other aspects, the north maintained species richness and community productivity equal to or above that of the open plots. This difference between north and south may be explained by plant shading (Table S1) that reduced evapotranspiration and heat stress in the north but not in the south (Butterfield et al., 2010;López-Pintor et al., 2006) rather than by rainfall interception as suggested by Tielbörger and Kadmon (2000). Thus, competition for water alone does not explain the lack of facilitation effects under reduced rainfall because these negative effects were ameliorated by reduced heat stress in the north. These results prove a failure of facilitation to maintain species that require obligate facilitation-except in the north aspect-as water became severely limited.

| CONCLUSIONS
Our results indicate that the effect of Retama nurse plants on herbaceous communities shifts from facilitative to competitive or interfering with decreasing water availability in support of the recent modifications to the stress-gradient hypothesis. Specifically, the balance between positive and negative effects by the nurse plant on the beneficiary community was spatially explicit shifting to competition in the east and the south and partially in the west-where diversity and productivity was lower under Retama relative to open locations.
However, the north aspect under Retama maintained more species and plants and supported a distinct community composition relative to open locations. Because nurse plants provide refugia for distinct species in these semiarid ecosystems, a climate that is changing to hotter and drier conditions and promoting shifts from positive to negative interactions will hinder the persistence of species that depend on nurse shrubs and in turn put biodiversity in these important ecosystems at risk.

DATA ACCESSIBILITY
All data used in this manuscript will be archived in Dryad.

CONFLICT OF INTEREST
None declared.

AUTHOR CONTRIBUTIONS
MOB analyzed and wrote the manuscript and compiled the data for the meta-analysis. FIP contributed to the writing and conceptual development of the manuscript. CA and SRE provided data for the meta-analysis and revisions to the manuscript. CS conceived the experiment, performed the data collection in 2012, contributed to the analysis, and cowrote the manuscript.