Environmental change makes robust ecological networks fragile

Introduction

Ecosystems, whether coral reefs, rain forests or a microbiome, are complex. How complexity evolves and its robustness to change is both a mystery to ecologists and a challenge for conservation biologists. To assess the current biodiversity crisis during global change requires looking beyond endangered species lists to extinction cascades1. Secondary extinction risk should be highest for specialists, an example being the loss of the condor louse after the condor became extinct in the wild2. In some circumstances, secondary extinctions could trigger even more extinctions up a food chain, unravelling entire ecosystems3. For this reason, the already appreciable endangered species list might be just the beginning4.

Although evolution should lead to both specialization and robustness to species loss, global change and human activity might change species vulnerabilities5,6, which could then decrease resource–consumer network stability. We investigated this hypothesis by contrasting how historical and novel conditions affected parasite assemblage robustness using computer simulations and information from global host–parasite databases. Ecological networks were fragile under environmental change due to the tradeoff between adapting to a stable past or an uncertain future.

Parasite assemblages are a convenient model for studying network robustness because they provide a straightforward, unidirectional response to host species loss (that is, host extinction, in general, affects parasite persistence but not the other way around). To estimate parasite assemblage robustness, one can apply analytical models, but it is more accurate to take a food web or a bipartite host–parasite network and then remove hosts in sequence (called host disassembly), recording the rate at which parasite richness declines7. Parasite robustness declines faster with increasing life-cycle complexity and parasite specificity7. However, parasite specificity is not independent from host extinction order. For instance, fish parasites tend to be either generalists or they specialize on ‘dependable’ hosts that are less vulnerable to extinction8. A similar pattern occurs within local food webs, with host/parasite networks being more robust to rare host removal than to random host removal7. However, current host vulnerability to extinction, measured by modern threats (for example, habitat destruction), can differ from historical vulnerability to extinction9, suggesting that dependable hosts in the past might not be dependable in the future5,6,10.

Here, we use artificial life simulations and empirical data to investigate how stable ecosystems would respond to extinctions under different scenarios of species loss. We show that ecosystems evolve complexity that is robust to historical conditions. However, under changing conditions, including current anthropogenic threats to biodiversity, robustness to change decreases, suggesting that future species losses should trigger secondary extinctions and eventual ecosystem collapse.

Results

Evolving digital host–parasite networks

We ran robustness experiments with digital hosts and parasites that self-replicate, mutate and compete on the artificial life platform Avida11. We chose different random settings covering various environmental scenarios (see the ‘Methods’ section and Supplementary Table 1) for each of the 100 evolution simulations. Each simulation started from a single host ancestor that ‘speciated’ with time. After a random number of host generations, we injected a random number of identical parasite individuals. As hosts and parasites diversified, host species varied in their vulnerability to extinction, and parasite species varied in their virulence, that is, in the percentage of central processing unit (CPU) cycles subtracted from a host. Because carrying capacity (that is, number of available hosts) remained constant throughout the simulation, the increased host and parasite diversity resulted in an average increment in specialization of interactions (Fig. 1a). Although specialization is expected to increase co-extinction risk, parasite assemblages gained robustness over time, reaching a maximum within about 5 × 104 generations (Fig. 1b). We continued the simulations until 1–5 × 105 generations, obtaining robust digital host–parasite interaction networks at different stages of maturity.

Figure 1: Robustness evolves rapidly in Avida even while specialization increases.

Robustness (a) was measured as the area under the curve of parasite diversity versus host diversity when a network was disassembled according to host historical vulnerability (see the ‘Methods’ section). Specialization (b) was computed as 1 minus the average fraction of hosts used by parasites in a given experiment. Solid lines and dots indicate average values. The Spearman-rank correlation coefficient was computed on all raw data (that is, not on the averaged values shown in the plots). Boxes indicate first and third quartiles, whiskers indicate range values and horizontal lines indicate median values.