Discussion
The development of novel high-throughput molecular methods has paved the way for a better understanding of how environmental factors shape the structure of species interaction networks (Nielsen et al. 2018; Roslin et al. 2019). Using an improved and non-invasive metabarcoding procedure based on insect feces, we reconstructed the structure of plant–orthoptera networks across multiple sites along elevation gradients, thus favoring technical advances in the field of ‘landscape network ecology’. We showed that networks exhibited structural variation along the ecological gradients, as a result of both the rewiring of species interactions and shifts in network size. Networks of high-elevation cold environments displayed reduced levels of specialization and increased nestedness, which are metrics theoretically expected to be central components of network stability (Fig. 2, Fig. 4; Table 1; Appendix S1, Fig. S4a, Table S1). We argue that when increased, these structural properties confer higher network resilience, presumably through a more homogeneous distribution of the herbivore interaction over the available plant species functional space. Theoretical work on the structure–stability relationship of ecological networks suggests a positive association between network resilience to species extinction and structural indices, including connectance and nestedness (Dunneet al. 2002; Memmott et al. 2004; Lafferty & Kuris 2009). Our empirical analyses along several elevation transects support theoretical expectations showing that networks in cold environments are less specialized and more nested, which presumably enhance network robustness (Burgos et al. 2007; Miller-Struttmann & Galen 2014). Novel molecular methods enabling the monitoring of network variation in space, as done in our study, but also in time, should provide new perspectives for understanding the trophic architecture of species assemblages.
The observed higher generality for alpine plant–orthoptera networks agrees with three underlying arguments supporting biotic and abiotic shifts along elevation gradients: (i) lower environmental predictability, (ii) less species competition for resources, and (iii) relaxation of plant chemical defences (Macarthur & Levins 1967; Hodkinson 2005; Rasmann et al. 2014). First, greater environmental stresses and variation in the alpine belt (Körner 2003; Barry 2008) may impose constraints for insects to complete developmental and reproduction cycles (Hodkinson 2005). In particular, environmental fluctuation at high elevations may increase resource stochasticity, which translates into greater spatio-temporal variation of the host plants than at low elevations (Billings & Mooney 1968). Cooler and more variable temperatures might also reduce search and digestive efficiency in ectothermic animals (Hodkinson 2005). In turn, such environmental unpredictably could be offset through the reinforcement of generalist feeding behavior (Macarthur & Levins 1967). Overall, orthoptera are generalist feeders (the median number of host plants in our study was 26). Hence, while food plant fluctuation should largely impact the evolutionary specialization of more specialized clades, such as the butterflies (median of eight host plants, Pellissier et al.2012b), orthoptera might more easily compensate for the demographic fluctuations of food plant species by maintaining a large diet breadth (Cates 1981) and other factors could contribute to the observed increase in generality. Second, higher species richness of orthoptera at low elevations (Appendix S1, Fig. S5) might pressure species to escape competition by focusing on distinct and more specialized diets (Macarthur & Levins 1967; Hodkinson 2005). However, we found no relationship between the species richness of orthoptera and the overall network specialization (Appendix S1, Fig. S6) and species niche overlap among orthoptera did not vary along the temperature gradient, indicating that interspecific competition for plant resources is weakened in orthoptera (Appendix S1, Fig. S4c, Table S1). Third, it was previously shown that alpine plant communities are less resistant to herbivores than low-elevation plant communities (Rasmann et al. 2014; Callis-Duehl et al. 2017). These plant defence patterns could promote a stronger generalist feeding behavior in colder environments, through easier digestibility of various plant materials (Moreiraet al. 2018). Our results indicating lower selectiveness of orthoptera for alpine plants are in agreement with a generalized reduction in defence levels in plants growing at high elevations (Rasmann et al. 2014). We documented that orthopteran communities from cold environments feed on a broader range of plant families and target more intensively families such as Apiaceae, Boraginaceae, Caryophyllaceae and Fabaceae, compared with the feeding habits of lower-elevation orthoptera (Appendix S1, Fig. S7). In particular, we found that some species at higher elevation were extreme generalist, feeding on almost all plant species in the communities (Fig. 4). The presence of those hyper-generalists might contribute to the reported increase of nestedness with elevation (Fig. 2c, Table 1), where, more specialist species interact more often with a proper subset of the most generalist species (Bascompte et al. 2003; Miller-Struttmann & Galen 2014). Russo et al. (2019) also showed higher nestedness as a result of a prevalence of super-generalist feeders and resources. Altogether, higher generality at the network and species level should increase the robustness of networks to extinctions.
The increase in generality and nestedness of networks at higher elevation (lower temperatures) were associated with an increase in network robustness in cold environments (Fig. 2, Table 1). These results support previously documented co-variation between network robustness and generality (Welti et al. 2017) or nestedness (Araújo 2016) in plant–herbivores systems, and a negative association between temperature and network robustness (Welti et al. 2019). The association between network specialization, nestedness and robustness metrics with the underlying temperature gradient suggests that ecological or evolutionary factors have led to more robust networks in more stressful environments. Consistently, increased generality and nestedness at high elevation may results from a greater number of plants species involved in realized interactions thus decreasing the weight of the keystones species (Fig. 3; Appendix S1, Table S2). In general, orthoptera feed on multiple plant species, so the loss of one plant species is never sufficient to cause the loss of one species of orthoptera (averaged keystone score <1), but they still show some degree of preference as regard to the functional traits of the plant they are feeding on. In our study, orthoptera showed a preference for plants with tougher leaves (Fig. 3) which typically correspond to monocotyledons (Fig. 1; Appendix 1, Fig. S8, Table S2), some of which were particularly dominant in the studied grasslands (e.g. Bromus, Festuca and Nardus ), as these herbivores are equipped with enough mandibular strength to cut through such leaves (Ibanez et al. 2013a). We found lower and more even keystone scores for alpine plant species, meaning that the removal of plant species at higher elevations was associated with lower secondary extinctions (Table 3). The decrease in the keystone score of grasses at higher sites might also be associated with the decline in the cover of grass vegetation with increasing elevation (Appendix S1, Fig. S9). At high-elevation sites, keystone species also had other functional attributes, including higher SLA but lower C/N values (Fig. 3) compared with low-elevation plants, corresponding to more palatable and resource-rich host plants (Pérez-Harguindeguy et al. 2013), providing herbivores with higher nutritive content during the short growing season of the alpine environment. These results suggest that the identity of the keystone species in plant–orthoptera bipartite systems is determined by a combination of factors involving plant species abundances and co-evolutionary mechanisms between insect feeding ability and plant defence, presumably resulting from mechanical and chemical defence tradeoffs.
Compared with traditional methods based on visual analyses of feces or gut content or literature-based documentation of interactions (Nielsenet al. 2018), the DNA metabarcoding procedure represents an effective and easily adaptable method for documenting interactions involving plants and insect species. As a compromise between the spatial coverage of our study and the available sampling resources, potential impacts of sampling replication and seasonality and year-to-year change on diet composition were not assessed here (Mata et al. 2019). Overall, our approach may open fields of investigation on the possible spatio-temporal variation in plant–insect interactions by expanding the means for collecting species interaction data.
Taken together, our results show an increase in plant–herbivore network generality and nestedness with elevation, which drives variation in network robustness along the gradient and ultimately gives lower weights to keystone species in alpine than in lowland environments. Shifts in abiotic components can alter the structuring of species interactions directly or indirectly (Welti & Joern 2015; Tylianakis & Morris 2017), by influencing the different aspects of the species interface through both abiotic and biotic pressures. We suggest that the observed patterns of network structural variation regarding elevation represent entangled responses of networks to environmental predictability and plant chemical defences, although further investigation would be required to confirm this possibility. Generally, orthoptera are not very sensitive to extinction, in that the loss of multiple plant species is necessary to cause secondary extinctions. Nevertheless, intensification of land use practices in lower-elevation mountain grasslands, for instance through the use of fertilizers, can regularly cause the loss of multiple plant species, which could then lead to extinctions in orthoptera (Chistéet al. 2016). Our study helps paving the way to a better understanding of the eco-evolutionary factors underlying network structure along large-scale ecological gradients, but also highlights how resilient species assemblages are to the accelerated rate of species extinction.