Introduction
The factors driving global scale patterns in ecological community structure are becoming increasingly well-known, with elevation, latitude and anthropogenic disturbances playing major roles (Nash et al In press; Romero et al 2020). However, because all organisms within an ecosystem are closely interconnected, variation in the distribution of a particular species in relation to the environment can also affect the species with which it interacts (Forister et al., 2012). In particular, a mechanistic understanding of how human activities alter the structure of ecological networks would be timely. This is because our world is facing accelerated anthropogenic changes, such as global warming, which causes species to move to higher latitudes and elevations (Franco et al., 2006; Memmott et al., 2007), and habitat disturbance, resulting in global diversity declines (Newbold et al., 2015). This variation in community structure is reflected in the structure of interaction networks, with consequences for multiple vital ecosystem services such as pest control (Macfadyen et al., 2011), fish stock production (Moi et al. 2022) and pollination (Burkle & Alarcón, 2011). Furthermore, current network structure predicts ongoing robustness to future natural and human-induced perturbations (Morton et al., 2022).
One of the most fundamental parameters used to quantify variation in network structure is ecological network specialisation (Dormann et al., 2008; Bascompte et al., 2003). The interaction between two groups can range from specialized, where a species interacts with a small subset of available partners and is more vulnerable to change and extinction, to generalized, where a species is less discriminate and interacts with a wide range of partners and is therefore predicted to be more resilient to change (Futuyma & Moreno., 1988).Most studies of network specialisation fall into one of two broad categories: (i) Those that compare different network types, such as different guilds, taxa, or trophic levels without accounting for environmental variation (Blüthgen et al., 2007; Cagnolo & Tavella, 2015, Guimarães et al., 2007), and (ii) Those that examine specialisation in a specific network type across environmental gradients, for example in relation to anthropogenic habitat disturbances or latitude (Schleuning et al., 2012; Gorostiague et al., 2023; Olesen & Jordano 2003; Luna et al., 2022). Studies from the former category generally reach consistent conclusions, for example, that pollinator networks are more specialized than seed dispersal networks (Blüthgen et al., 2007). However, in the latter category results are often inconsistent between studies, sometimes depending on network size, taxonomic scope and/or geographical range of networks employed. For example, some plant-pollinator systems are more specialized in the tropics (Gorostiague et al., 2023), while others have been found to be more specialized in temperate regions (Schleuning 2012). The specialisation of plant pollinators on plant species has been reported to increase with elevation on islands (Olesen & Jordano, 2003), while a recent global analysis concluded that latitude and elevation did not play a role in explaining the degree of pollinator specialisation (Luna et al., 2022). As different network types can interact with environmental gradients in a complex way, the simultaneous evaluation of by drivers of specialisation is currently needed.
The application of different methods across different studies for measuring network specialisation, some of which may be confounded by other aspects of network structure (e.g. network size, network connectance), means that global-scale patterns in specialisation remain unclear (Pellissier et al., 2018). The classical method for measuring specialisation at the network level, the H2’ index, is based on the deviation from the expected probability distribution of interaction frequencies between species (Blüthgen et al., 2006). However, networks of different sizes, involving different numbers of species and links, can vary in the observed value of H2’ regardless of specialisation (Blüthgen et al., 2008). One way to overcome this shortcoming is to compare observed values of H2’ to the distribution resulting from repeated randomisation of the original network (Dormann at al., 2009). This approach can generate a standardised effect size that is less affected by other aspects of network structure (H2’ z-score; Ulrich et al., 2009). Specialisation can also be affected by availability of partner species, in particular those that are closely phylogenetically related (Segar et al., 2020). The distance specialisation index (dsi*) allows incorporation of phylogenetic relationships and species abundances when quantifying specialisation (Jorge et al., 2014; 2017). However, the degree to which phylogenetic relatedness of resources determines consumer specialisation in networks at large spatial scales is currently unknown. Such phylogenetic signals in the interactions between species within the network are likely the result of reciprocal coevolution or long-term adaptation of consumer species to traits of resource species (Guimaraes et al, 2007). For example, common milkweed (Asclepias syriaca ) produces chemical defensive compounds that only a small number of closely related herbivorous beetle can tolerate (Rasmann & Agrawal, 2011). A specialist may interact with a suite of closely related resource species because these are more likely to have the traits to which the consumer has adapted (Rasmann & Agrawal, 2011). Hence there is a need to explore how specialisation changed over the gradient of latitude and elevation, and the presence of disturbance, as well as how phylogenetic relatedness of the consumer or host influences these patterns.
Interactions between ants and plants provide a suitable system for exploring changes in network specialisation along environmental gradients. Ant-plant interactions are widespread and involve discrete interaction types with varying degrees of specialisation (Ness & Lach, 2010) that are easily categorised. Furthermore, ant-plant systems are well-studied, making a data collation and meta-analytical approach tenable. Finally, there are well-resolved phylogenies for both groups, allowing exploration of degree of phylogenetic specialisation in networks. Ant-plant interactions range from obligate symbiotic mutualisms such as those between myrmecophytic plants and their long-term ant partners, to non-symbiotic facultative interactions such as myrmecophily and myrmecochory (Heil & McKey, 2003), to less specific interactions in which ants opportunistically forage on plants (Rico-Gray & Oliveira, 2007). We define the myrmecophytic interactions as those in which domatia are formed on plants to provide shelter for ants. Myrmecophilic interactions, are those in which extrafloral nectaries and fruiting bodies are produced by plants as a food source for ants, but plants do not provide domatia as nest sites. Note that some domatia bearing plants also provide extrafloral nectar and food bodies, but we nonetheless classify these as myrmecophytes. Myrmecochorous interactions are those in which ants help plants disperse seeds in exchange for food (Heil & McKey, 2003). Finally, plants can provide a space for foraging and patrolling without any apparent evolved mutualism, which we define here as “foraging” interactions (Rico-Gray & Oliveira, 2007).
Although the specialisation drivers in local ant-plant networks have been studied extensively (Juárez-Juárez et al., 2023) commonality of patterns at global scales remains unclear. Myrmecophytic interactions tend to be more specialized than other kinds of ant-plant interactions, such as myrmecophily, seed dispersal (Blüthgen et al., 2007), and frugivory (Guimaraes et al., 2007), presumably driven by tighter co-evolution due to the greater intimacy of the interaction (Pires & Guimaraes, 2013). These myrmecophytic networks become less specialised with increasing elevation (Plowman et al., 2017) and anthropogenic disturbance (Emer et al., 2013), probably due to loss of partner species. However, they appear to be robust to forest fragmentation (Passmore, 2012). Yet, it is currently unknown how specialisation of myrmecophytic networks varies with latitude. Interestingly, myrmecophilic networks show levels of specialisation similar to ant-lepidopteran networks (Cagnolo & Tavella, 2015), perhaps because ants are provided with carbohydrate-rich liquid food by plants and lepidopteran larvae respectively in these cases. Interactions in which ant workers forage on plants, but are not involved in mutualisms, are expected to show lower specialisation than other interaction types, although this has not been directly studied. The specialisation of such foraging networks does not change along a 20° range in latitude (Dáttilo & Vasconcelos 2019), although specialisation is reduced following anthropogenic forest disturbance (Corro et al., 2019). Hence, whether different types of interaction between these two ecologically important groups show common responses to latitude, elevation and anthropogenic disturbance remains unclear.
We collated globally distributed network data and used a meta-analytical approach to determine how the specialisation of ant-plant networks varies with interaction type (myrmecophytic, myrmecophilic, myrmecochorous, and foraging), anthropogenic disturbance, latitude, and elevation at global scales, including interactive effects between these predictors. We predicted that myrmecophytic interactions would be the most specialised as these involve tight co-evolutionary adaptations between partners, followed by myrmecophilic, myrmecochorous, and foraging. We also predicted that ant-plant specialisation will decrease with increasing latitude, elevation, and anthropogenic disturbance due to the scarcity of resources in these areas (Brown, 2014).