In an article published in Ecology and Evolution on 13 June, Strona and Veech demonstrate how the underlying assumption in ecological network analysis that all species interactions are possible (e.g. all plants can potentially be pollinated by a given insect, and all insects can potentially use a certain plant) can produce misleading results.
An understanding of the structure of ecological networks can help unravel the processes that generate and shape biodiversity, and tell us about the how ecological communities might react to different impacts, such as climate change, overexploitation, etc.
Several studies have investigated the relationship between the structure of ecological networks (such as the interactions between plants and pollinators) and their persistence. More specifically, ecologists have hypothesised that competition between two species for a limited resource could decrease when the number of shared mutualistic(1) partners increases.
By enhancing the population growth of the shared mutualistic species, the two competing species might have less of a negative effect on one another than they would otherwise. This type of interaction may reduce the overall competitive load within a network, perhaps making it more stable both in terms of species diversity, and the strength and sign of the interactions (Fig.1).Fig 1. If two competing species (e.g. two pollinating insect species) have a mutualistic relationship with the same partner (e.g. a flowering plant), they would both promote its population growth, which would result in a reciprocal positive effect for themselves (A). When there is more than one potential mutualistic partner available, a nested structure might minimise the overall competitive load (B, C), while other kinds of structure where competitors share no mutualistic partners (D) would have a high overall competitive load.
However, to demonstrate this hypothesis, it is crucial to assess whether, and to what extent, species tend to share mutualistic partners. The common approach to do this consists in comparing the observed degree of sharing with that expected in a situation of ‘random’ ecological interactions.
Take, for example, a plant-pollinator network. This would consist of two sets of nodes that represent, respectively, plant and insect species. Arrows connecting nodes that belong to the two sets would indicate which plants are pollinated by which insects.
Given this network, one may quantify how often, on average, two pollinators use the same plant, and how often, on average, two plants are visited by the same pollinator. These observed interactions can then be compared with those obtained by randomising the arrows connecting plants to pollinators.
If the observed degree of overlap is higher than that of most (95%) of the randomised networks, then it is usually concluded that the observed pattern is significant.
Although this procedure seems perfectly reasonable, there has been much debate around the use of randomised networks to determine whether or not there is a tendency to share partners, mostly revolving around how the procedure should ensure the retention of various features of the original network (such as, for example, the number of plants and insects, the number of interactions, the number of plants pollinated by a given insect, the number of insects visiting a plant, etc.).
However, little attention has been paid to the fact that all of the randomization approaches are based on the underlying assumption that all interactions between species are possible (e.g. all plants can be potentially pollinated by a given insect, and that all insects can potentially use a certain plant in a pollination network).
In their article, Strona and Veech demonstrate how this assumption, which clearly lacks ecological realism, can produce misleading results. In all kinds of ecological networks, a variety of constraints limit the set of potential partners available for a given species. Sticking to the abovementioned example, the ability of insects to use a certain flower depends on the morphology of their mouths, which, in turn, is the result of a long history of co-evolution between the pollinator and its host plants.
Similarly, in a food-web, there are many constraints that limit the range of a predator's potential prey. For fish, for example, a prey must almost always be smaller than its predators. Now, assume that two different pollinators in a large network are observed visiting only one (the same) plant species.
It would clearly be important to understand whether the target plant is the only species accessible to the two pollinators, or if the overlap in resource use should be attributed to other reasons potentially of greater ecological interest (such as some beneficial interaction between the two pollinators). In their 2015 article, Strona and Veech introduced a metric that they now show well suited to this goal.
By applying their metric to artificial simulations and real data, Strona and Veech discuss and illustrate how it is of paramount importance to identify the set of ‘permitted’ interactions for a given species (i.e. interactions that are not impeded by, for example, a lack of functional trait compatibility) in order to understand the ecological and co-evolutionary processes at the basis of the network structure.
They also show how the choice of different criteria may yield substantially different outcomes. These results suggest that a more conscious attempt to take into account permitted vs. forbidden interactions could be the key to a more complete understanding of patterns and processes in ecological networks.
(1) a relationship between two species of organisms in which both benefit from the association