Introduction
Species represent the main building blocks of ecosystems and are connected in webs of positive and negative interactions, which shape ecosystem processes and functioning (Thompson et al. 2012). Given the central role of interactions among species for energy and matter flow between ecosystem compartments (Barnes et al. 2018), studying the structure of ecological networks helps us understand how ecosystem functioning might be disrupted by global changes (Petcheyet al. 1999; Tylianakis et al. 2008). The wiring among interacting species is hardly random but rather governed by ecological rules (Bascompte 2010; Laigle et al. 2018). The strength of interactions between species may depend on the degree of matching between functional traits, which are shaped through co-evolutionary processes (Rausher 2001; Laigle et al. 2018). In turn, rules of functional matching might be influenced by variation in environmental conditions, such as temperature (Sentis et al. 2014; Gounandet al. 2016), or by climatic stability (Dalsgaard et al.2011). By inducing changes in species composition, ecological gradients can be associated with shifts in species co-occurrence and their ability to form stable links (Welti & Joern 2015; Pellissier et al.2018). Moreover, shifting environmental conditions might influence interactions among species even when they are steadily co-occurring (Tylianakis & Morris 2017). As a result, interactions shifts along climatic clines can lead to changes in the structure of networks (Welti & Joern 2015). Nevertheless, the geographic variation in networks is poorly studied, owing to the difficulty of documenting multiple ecological networks along climatic gradients.
There are major challenges to the large-scale study of ecological networks that relate to the documentation of interactions and the methods used to perform network comparisons at the landscape scale (Pellissier et al. 2018). The study of ecological networks along environmental gradients has so far been limited by the difficulty of observing comparable interactions simultaneously at multiple locations. Novel DNA metabarcoding methods, which are increasingly cheaper, faster and more comprehensive, have opened such opportunities (Kaartinenet al. 2010; Roslin et al. 2019). Deagle et al.(2007) were among the first to develop a DNA metabarcoding protocol to reconstruct the trophic regime of the macaroni penguins on Heard Island in the Indian Ocean. Since then, the study of entire ecological networks has been facilitated through the adaptation of DNA metabarcoding techniques to different sample sources, which enables the collection of many samples over a short period of time (Roslin et al. 2019). For instance, Pornon et al. (2016) developed a protocol to quantify plant–pollinator interactions from pollen samples in the French Central Pyrenees, while Ibanez et al. (2013) applied this approach to insect feces for studying the diet of insect herbivores. However, most protocols for network reconstruction were not designed for studies with large spatial scales, and not all were aimed at species-level resolution. In addition, a complete description of the wet lab and bioinformatic procedures is not always accessible, limiting the adaptability and reproducibility of the techniques used to document species interactions. Scaling up the utilization of DNA metabarcoding to entire landscapes, while also sharing methodological workflows as detailed and user-friendly protocols, can spur advances in the study of species interactions along environmental gradients.
From the wide range of natural gradients impacting species distribution and interaction patterns, montane clines represent optimal natural laboratories to understand how species and their interactions vary over environmental gradients (Körner 2003). Changes in climate – most notably temperature – along elevation gradients cause strong environmental filtering in communities (Rahbek 1995; Hodkinson 2005) and can therefore also be expected to influence the structure of ecological networks (O’Connor et al. 2009; Welti & Joern 2015). The structure of ecological networks along environmental gradients can change as a result of two main processes: (i) a turnover of the species in the network, or (ii) a turnover of the links in the network, in which co-occurring species rewire their interactions along the gradient (Gravel et al. 2019). In particular, the steady decrease in temperature with increasing elevation has been associated with changes in species richness and abundance (Rahbek 1995; Hodkinson 2005; Descombes et al. 2017), likely influencing the networks of species interactions (Adedoja et al. 2018; Pellissier et al. 2018). Therefore, studying changes in the architecture of species networks along elevation gradients contributes to evaluations of the effect of temperature on community structure and stability.
Changes in species’ interactions within networks can be summarized by a set of indicators relating to the degree of structuration and complexity of the network (Delmas et al. 2019), including connectance (Martinez 1992), generality (Bersier et al. 2002), nestedness (Bascompte et al. 2003) and robustness (Dunne et al.2002). Metrics of network structure can also quantify the resilience of the networks to environmental disturbances (Thebault & Fontaine 2010). Specialized networks have been found to be associated with lower robustness against species extinction (Lafferty & Kuris 2009; Tylianakis & Morris 2017; but see May 1973; McCann 2000). This is the result of the existence of keystone species (Paine 1969), which are nodes of interactions on which relies the system stability (Poweret al. 1996). The importance and identity of keystone species, but also general structural properties involved in network resilience, may reshuffle along elevation clines. Three main non-exclusive hypotheses have been proposed to support this pattern: (i) at higher elevations the environment is expected to be less predictable (Barry 2008), and survival under these conditions necessitates the evolution of a broader diet breadth (Macarthur & Levins 1967); (ii) more intense competition at low elevations is predicted to select for more specialized diets to decrease niche overlap (Macarthur & Levins 1967; Hodkinson 2005); and, more closely linked to plant–herbivore interactions, (iii) a decline in the capacity of plants to resist herbivore attack at higher elevations is expected to facilitate a larger diet breadth of herbivores (Pellissier et al. 2012a; Rasmannet al. 2014; Moreira et al. 2018). In contrast to the expectation of increased herbivore generality at higher elevations, where plant communities are less diverse, it has been proposed that higher plant species richness could benefit insect generalists simply by increasing the availability of species to feed on (Unsicker et al. 2008, Welti et al. 2017).
The comparison of ecological networks along environmental gradients has the inherent methodological difficulties of network comparison (Pellissier et al. 2018). Effective analyses of network structure isolate the influence of real interaction patterns on network structural indices from the effects of network size or sampling design (Banašek-Richter et al. 2004). Studying ecological networks along large-scale environmental clines is challenging in this regard, as the decline in species richness at the extremes of the gradient might result in significant variation in species richness, in turn affecting measures of structural indices sensitive to matrix size (Trøjelsgaard & Olesen 2016; Pellissier et al. 2018). For instance, the number of links per species inevitably declines from large to small networks, as larger networks include more possible links. Several strategies have been developed to alleviate the confounding effects of comparing networks of different sizes (see Pellissier et al. 2018 for a review of these approaches). The most commonly used approach is the null model, where networks of randomly distributed interactions are generated and compared with the empirical patterns. This method has been established to isolate the role of observed interaction patterns on the network structure from the effect of matrix size variation when comparing networks along environmental gradients (Vázquez & Aizen 2003).
In this study, we investigated the variation in the structure of plant–orthoptera ecological networks along elevation gradients. Orthoptera are among the most abundant herbivorous arthropods in semi-natural grasslands of the European Alpine system, and they strongly impact the functioning of these ecosystems (Blumer & Diemer 1996). We optimized a protocol for plant DNA metabarcoding applied to orthopteran feces in order to reconstruct plant–orthoptera bipartite networks across 48 study sites situated along six elevation transects in the Swiss Alps. We then applied null models to explore the structural variation in plant–orthoptera bipartite networks and determine if lower levels of network organization and increased robustness are associated with the low temperatures of the alpine environment. Specifically, we proposed the following three hypotheses:
  1. Levels of generality in orthoptera should increase in colder environment (higher elevation) according to the following lines of argument: generalist feeders are better equipped to compensate for higher environmental uncertainty; lower interspecific competition attenuates positive selection for specialization; and the reduced plant chemical defences typically found at higher elevations offer more dietary opportunities for insects.
  2. The robustness of the network after simulated plant primary extinctions should be higher in colder environments (at high elevations). The increase in insect generalist feeding behavior predicted in the first hypothesis should allow networks of alpine communities to better compensate for possible plant species loss.
  3. If more generalist insect herbivores are present and the network is more robust in colder environment, the removal of plant species should induce fewer extinctions within orthopteran communities than at low elevation so that plants have lower keystone weights.