DISCUSSION
The dominant shrub A. ordosica had an overall neutral effect on the biomass and richness of herbaceous understorey species in the controlled conditions of this dune system, and neither watering nor N fertilization significantly influenced the effect of the shrub on herbaceous species at the community level. However, within the community, we identified contrasting changes in species responses to the effect of the shrub along watering and N fertilization treatments that were related to the two measured traits, height and LDMC. When species were grouped based on the differences in both their traits and responses to the effects of the shrub, we found significantly contrasting changes in species group-level responses to the effects of the shrub with watering and N fertilization, that balanced at the community level, consistent to our main hypothesis. Our results have strong implications, for understanding both changing in species responses to neighbours along environmental gradients, a highly debated issue in community ecology, and the stabilizing effects of functional divergence and plant-plant interactions for community responses to global change.
In our experiment, four species groups could be clearly separated according to their divergences in functional traits (height and LDMC) and their changes in responses to the effects of the shrub with watering and N fertilization. These results support former theories arguing that plant communities include different functional groups of species having different ecological requirements and plant traits (Ellenberg, 1953; Whittaker, 1956; Lavorel et al. , 1997). They are also consistent with results of pioneer climate change experiments that showed that climate change differently affects species of contrasting functional strategies within a single community (Chapin & Shaver, 1985; Harte & Shaw, 1995). However, to our knowledge, this is the first evidence that environmental manipulations mimicking global change effects in a single community differently affect the responses to a dominant neighbour of species having different functional trait values. Plant height appeared to be the best predictor of plant responses to the effect of the shrub, with the tallest species (group C) being facilitated and the shortest ones (groups A and B) overall negatively affected by the shrub. These findings are consistent with previous studies showing that competitive responses are more related to plant height than LDMC (Michalet, 2001; Liancourt et al. , 2005b; Michalet et al. , 2008; Wang et al. , 2019b). Additionally, Liancourt et al. (2005a) have shown that short stress-tolerant species are more likely to be negatively affected by neighbours than tall competitive species, since the benefit of being shaded by neighbours for mitigating physical stress outweighs the cost for light for the latter but not the former. Although significant, relations between LDMC and responses to neighbours along the treatments were more complex to understand. For example, group D including the species with the highest LDMC values (i.e., being the most conservative; Díaz et al. , 2004 & 2016; Wright et al. , 2004), was strongly negatively affected by the shrub in the W40N0 treatment, but facilitated in W40N60 treatment, which response is opposite to those of group C. This might be explained by indirect interactions among different groups of competitors having contrasting resource requirements, as shown by Michalet et al. (2015) who assessed direct and indirect interactions at the species level in a subalpine shrub community from the Tibet Plateau.
Interestingly, changes in the effects of the shrub along the watering and N fertilization treatments supported different predictions of competition and facilitation models, depending on the species cluster groups and resources. For group A, facilitation decreased and competition increased with increasing N fertilization, supporting the SGH model (Bertness & Callaway, 1994). For group B, RII did not change significantly with neither watering nor N fertilization, a result supporting the meta-analysis of Maestre et al. (2005) for a drought gradient and the model of Tilman (1982) for a nutrient gradient, respectively. For group C, facilitation was the highest in the wettest conditions and decreased with increasing drought, with no significant competition or facilitation in the driest conditions, which result supports the collapse of facilitation model (Michalet et al. , 2006). Finally, for group D, competition increased with watering in nutrient-poor conditions, but facilitation switched back to competition with increasing drought in nutrient-rich conditions, results supporting the SGH and switchback to competition models, respectively (Bertness & Callaway, 1994; Maestre & Cortina, 2004; Michalet et al. , 2014a). Together these results suggest that apparent inconsistencies in results of experiments on variation in competition and facilitation along environmental stress gradients might be explained by the functional strategies of the target species involved in the different studies or the type of stress gradient (Michalet, 2007; Maestre et al. , 2009; Michalet et al. , 2014a; Liancourt et al. , 2017) rather than methodological issues as argued by He & Bertness (2014).
Consistent to our main hypothesis, the contrasting between-group variations in the effects of the shrub along the watering and N fertilization treatments balanced at the community level, since there were no significant effects of the treatments on RII for community biomass and richness. Previous studies have found community-level balance of the effects of dominant neighbours among groups of species within a community (Michaletet al. , 2015; Wang et al. , 2017). These authors have argued that this is more likely to occur in species-rich communities of intermediate stress conditions. Indeed, in extreme conditions of low or high stress, functional divergence is thought to be low, consistent to the humped-back model of Grime (1973) and, thus, communities are dominated by either competitive or stress-tolerant species with a negative or positive net response to neighbours at the community level (Michalet et al. , 2006). However, to our knowledge, this is the first experiment showing that significant changes in the effects of a dominant shrub along environmental treatments mimicking global change balance at the community level. These new findings are of crucial importance for including biotic interactions in predictive models of species and community responses to global change, a current important goal in global change research (Tylianakis et al. , 2008; Wiszet al. , 2013; Anthelme et al. , 2014; Michalet et al. , 2014b; Svenning et al. , 2014). It shows that, although global change may have strong significant effects in plant communities at species level, increasing or decreasing competition or facilitation depending on the species functional strategy and stress factor involved, there could be no significant changes in the effects of dominant neighbours on community biomass and richness. We argue that this is more likely to occur in species-rich communities form intermediate stress conditions that exhibit a high functional divergence, as proposed by Grime (1973) and Michalet et al. (2006) and shown by a number of studies (Gerhold et al. , 2013; Duru et al. , 2014; Fryet al. , 2018; but see de Bello et al. , 2006). Thus, species diversity, and in particular functional richness, might be considered as insurance for community resistance to global change, and functional divergence and plant-plant interactions as stabilizing factors for community response to global change. These results do not imply that global change should not affect community composition when community-level changes in interactions are not significant, since there could be important changes at the within-community patch scale. For example, from our results, we can expect that some species will increase in abundance or biomass in open patches relatively to below shrubs when N deposition increases (case of group A in our experiment and stress-tolerant and ruderal species in general), and that other species will increase conversely below shrubs relatively to open patches when rainfall increases (case of group C in our experiment and competitive species in general). Such internal reorganization of the community in response to global change has already been observed in communities with strong environmental heterogeneity such as European calcareous grasslands (Fridleyet al. , 2011). These results also suggest that disturbing communities through the removal of dominant species may certainly disrupt community stability. Indeed, this could alter these contrasting interactions among functional groups and decrease community resistance to global changes as shown with modelling by Losapio & Schöb (2017) for alpine cushion-plant communities.
To conclude, our study showed that species of contrasting functional trait values within a single community exhibit contrasting changes in responses to the effect of a dominant shrub along environmental treatments mimicking global change, and that these interactions balance at the community level with no effects for the biomass and richness of the community. These results suggest that functional divergence and plant-plant interactions stabilize community response to global change. Our results are crucial for understanding the mediating role of plant-plant interactions for species and community response to global change and may help reconciling the conflicting literature on variation in facilitation and competition along environmental gradients.