Impacts of climate change on individual species are increasingly well documented, but we lack understanding of how these effects propagate through ecological communities. Here we combine species distribution models with ecological network analyses to test potential impacts of climate change on >700 plant and animal species in pollination and seed-dispersal networks from central Europe. We discover that animal species that interact with a low diversity of plant species have narrow climatic niches and are most vulnerable to climate change. In contrast, biotic specialization of plants is not related to climatic niche breadth and vulnerability. A simulation model incorporating different scenarios of species coextinction and capacities for partner switches shows that projected plant extinctions under climate change are more likely to trigger animal coextinctions than vice versa. This result demonstrates that impacts of climate change on biodiversity can be amplified via extinction cascades from plants to animals in ecological networks.

Climate change forces species either to move or to adapt to changing conditions1,2. Although models predicting the responses of individual species to climate change are widely utilized2, it is not yet clear to what extent a changing climate will affect biotic interactions between species3,4. Ecological theory predicts that abundant generalist species tend to have large ranges5 and, consequently, occupy wide climatic niches6, whereas species specialized on specific interaction partners have small ranges, occupy narrow climatic niches and may therefore be particularly vulnerable to climate change7.

In ecological communities, species are embedded in networks of interacting species, for instance, in mutualistic networks between plant species and animal pollinators or seed dispersers8. Species in these networks vary in the number of interaction partners, for example, because of differences in species traits9, and thus differ in their degree of biotic specialization8. So far, it has not been tested how biotic specialization in ecological networks relates to a species’ climatic niche breadth and its vulnerability to climate change. However, a quantitative understanding of this relationship is required to predict the likelihood of species extinctions and coextinctions from ecological communities under climate change10.

Here we test the two hypotheses that plants and animals with (1) narrow climatic niches and (2) a projected loss in climatic suitability are biotic specialists that interact with a low diversity of partners. We additionally simulate (3) how the relationship between biotic specialization and vulnerability to climate change affects the risk of species coextinctions of plants and animals under future climatic conditions. We analysed data on climatic niche breadth for 295 species of plants and their insect pollinators (196 bee, 70 butterfly and 97 hoverfly species) and seed dispersers (51 bird species) from central Europe. For each species, we quantified the change in climatic suitability across a species’ current European range under projected climate change according to two circulation models and two representative concentration pathways (RCPs 6.0 and 8.5). We linked projected changes in climatic suitability to data on biotic specialization derived from 8 quantitative pollination and 5 quantitative seed-dispersal networks recorded in 13 regions across central Europe. Networks describe interaction frequencies between plant and animal species, that is, the number of visits of an animal to a plant species, and yield empirical estimates of biotic specialization for each species in each network.

We find that animal species with narrow climatic niches and a projected loss in climatic suitability interact with a low diversity of plant partners, whereas we do not find analogous relationships for plants. This important difference between plant and animal species affects the likelihood of species coextinctions under climate change. We simulate different scenarios of species coextinction and capacities for partner switches and show with these simulations that mutualistic networks are more sensitive to projected plant than to animal extinctions under climate change. We conclude that a high potential for adaptive partner switches is required to stabilize mutualistic networks against extinction cascades from plants to animals under climate change.

Climatic niche breadth and biotic specialization

In line with the first hypothesis, we found that animals’ climatic niche breadth was positively associated with the effective number of plant partners in the regional pollination and seed-dispersal networks (Fig. 1a,b). In contrast, climatic niche breadth of plants was not related to the effective number of animal partners (Fig. 1a,b). For both plants and animals, we found no relationship between climatic niche breadth and complementary specialization d′ (a measure of the uniqueness of interaction partners relative to other species; Supplementary Table 1). These trends were qualitatively similar across the individual networks (Supplementary Table 2).

Figure 1: Biotic specialization in relation to climatic niche breadth and vulnerability to climate change.
Figure 1

Associations of (a,b) realized climatic niche breadth (climatic hypervolume60, OMI climatic niche breadth61) and (c,d) projected climatic suitability change (RCP 6.0, RCP 8.5 scenarios65; year 2070) with the effective number of partners (eH) of plant (n=295) and animal (n=414) species in 13 mutualistic interaction networks from central Europe. Specialization is the effective number of interaction partners66 of plant (blue) and animal (red) species in each network (shown on a log-scale). Trend lines indicate the estimated slope (β) in a mixed-effects model accounting for effects of network identity and animal and plant taxonomy on model intercepts. Shown are species’ mean partial residuals plus intercept from these models; symbol size is proportional to the weight of each species in the analysis, corresponding to its number of occurrences across networks and, in the case of climatic suitability change, the accuracy of the species distribution model (TSSmax value64); given are slope estimates±1 s.e. for plants and animals, P values were derived by Kenward–Roger approximation: **P<0.01 and ***P<0.001 (for full statistics see Supplementary Table 1).

Full size image

Vulnerability to climate change and biotic specialization

Consistent with the second hypothesis, we found that animals projected to lose climatic suitability across their current European range had a low diversity of plant partners in the regional networks (Fig. 1c,d). There was no analogous relationship for plants and their effective number of animal partners (Fig. 1c,d). Changes in climatic suitability were unrelated to complementary specialization d’ for both plants and animals (Supplementary Table 1). Trends were again similar across the individual networks (Supplementary Table 2).

We simulated secondary extinctions of animal and plant species from mutualistic networks as a consequence of sequential species loss from the other trophic level. Extending upon previous simulations of species coextinctions11,12,13, we informed our simulation model with projected changes in climatic suitability for plant and animal species and removed species sequentially from the highest to the lowest decrease in climatic suitability (Fig. 2). We modelled species coextinction under different scenarios of species’ sensitivity to partner loss assuming that a 25, 50 or 75% decrease in total interaction frequency would trigger the secondary extinction of a species from the regional network. These thresholds are probably more realistic than the assumption that all interaction partners must be lost to trigger secondary extinction11,12,13, given the frequency of coextinctions reported in empirical and modelling studies14,15. In the simulation, we further accounted for the potential flexibility of species in the choice of their interaction partners by reallocating a varying proportion of lost interaction events to persisting species. We account for a potential rewiring of interactions to new partners13,16 by comparing a scenario of constrained rewiring to persisting partners with a scenario of unconstrained rewiring to all persisting species. We did not consider, however, that new species may enter the interaction networks and, thus, may overestimate extinction risks under climate change17. Furthermore, we assumed that interaction frequencies indeed reflect the reciprocal functional dependences of animals on plants and vice versa18. For each network and simulation scenario, we quantified the relationship between primary and secondary species extinctions, yielding a measure of network sensitivity to plant and animal extinction, respectively (Fig. 2). We compared network sensitivity to species coextinction between extinction sequences due to climate change and due to random extinction, thereby accounting for effects that are independent of the extinction sequence, such as inherent differences in species numbers and mean specialization between plants and animals.

Figure 2: Secondary animal and plant extinction under climate change.
Figure 2

Shown are (a,b) secondary animal extinction in response to plant extinction and (c,d) secondary plant extinction in response to animal extinction for a seed-dispersal network from Białowieża forest (network ID=S1; 12 plant and 29 bird species). (a,c) Species (rectangles in red (animals) and blue (plants), connected by weighted interaction links; box and line width correspond to interaction frequencies) are removed sequentially according to projected suitability changes in climatic conditions. Low ranks (light shade) correspond to a high vulnerability to climate change, high ranks (dark shade) correspond to a low vulnerability; thus, light links are prone to extinction, whereas dark links are the persisting backbone of interactions under climate change. The corresponding secondary extinction plots (b) for animals (red) and (d) plants (blue) show network sensitivity to species extinction (filled area above the extinction curve) under four scenarios of species’ flexibility (solid to dotted lines) to reallocate interactions to persisting partners (constrained rewiring); here secondary extinction is triggered after 50% interaction loss. In this network, sensitivity to plant extinction (red area) was larger than sensitivity to animal extinction (blue area), that is, animal species went more quickly secondarily extinct than plant species. Secondary extinction plots for the 12 other interaction networks are shown in Supplementary Fig. 1.

Full size image

Across networks, we found that secondary animal extinctions were more likely to occur than secondary plant extinctions (Fig. 3 and Supplementary Fig. 1). In almost all scenarios, this difference was larger under climate change than under random extinction (Fig. 3) independent of the chosen RCP scenario (Supplementary Fig. 2). Differences between climate change and random extinction were most pronounced if we assumed a high species’ sensitivity to coextinction and a low capacity for rewiring lost interactions to other partners (see, for example, Fig. 3a). Scenarios in which many interactions needed to be lost to trigger secondary extinction differed less between climate change and random extinction, especially if species were able to reallocate many of their lost interactions to persisting species in the network (see, for example, Fig. 3f). Hence, animal coextinctions in response to plant extinction were most frequent if animals were limited in their flexibility to respond to future changes in partner availability.

Figure 3: Differences in sensitivity to species extinction across 13 mutualistic networks.
Figure 3

Shown are differences in network sensitivity to plant versus animal extinction for different scenarios of species’ sensitivity to coextinction, rewiring capacity and flexibility. Coextinction thresholds varied between (a,b) 25%, (c,d) 50% and (e,f) 75% of interaction loss. Species were able to rewire interactions (a,c,e) to persisting partners (constrained rewiring) or (b,d,f) to all persisting species in each network (unconstrained rewiring). Flexibility values (0%, 25%, 50%, 100%) indicate the proportion of lost interactions that was reallocated to other species in the respective scenario; we omitted the very unlikely scenario of unconstrained rewiring and 100% flexibility as it requires all species to go extinct to trigger secondary extinction. Shown are mean differences (±1 s.e.) across the 13 pollination and seed-dispersal networks between the impact of plant versus animal extinction; values >0 (red bars) indicate a higher risk of secondary animal than secondary plant extinction and values <0 (blue bars) indicate the opposite. Secondary animal versus secondary plant extinction was compared between climate change and random extinction using two-sided, pair-wise t-tests (+P<0.1; *P<0.05; **P<0.01). Here climatic projections of the models of species’ vulnerability to climate change follow the RCP 8.5 scenario; results were identical for the RCP 6.0 scenario (see Supplementary Fig. 2).

Full size image

For animals, but not for plants, our results support the first hypothesis that species with narrow climatic niches are biotic specialists. Different, not mutually exclusive, explanations are consistent with this finding. First, animal species with wide distribution ranges and climatic niches are usually locally abundant19 and are therefore likely to locally interact with more plant partners than rare species. In contrast, the relationship between range size and local abundance is usually more variable for plants20. Second, species traits that favour biotic generalization, for example, large body size21, may also favour the widespread distribution of animal species across a wide climatic range22. Thus, climatic niche breadth and biotic specialization may be indirectly linked via species traits. Third, realized climatic niche breadth and biotic specialization will be directly linked if the distribution of specialized animal species is constrained by that of their resource plants, which has been demonstrated for antagonistic plant–animal interactions of butterflies and other phytophagous insects7,23, but not yet for animal species linked to plants by mutualistic interactions. In contrast to animals, plants may depend less on their animal partners because pollination and seed dispersal by animals are characterized by a high degree of animal redundancy24 and because many plants have evolved alternative regeneration loops, such as clonal propagation, autonomous self-pollination or the maintenance of persistent seed banks25.

In line with our second hypothesis, we found that specialized animals may indeed be more vulnerable to climate change than generalists. Thus, climate change is likely to trigger a decline or even the local extinction of animal species that are constrained by the occurrence of specific plant partners7,23. However, as the most connected animals seem to be relatively tolerant to projected changes in climatic conditions, climate change may only have weak indirect impacts on ecological networks via top-down effects from animals to plants. In contrast to animals, we did not detect a relationship between biotic specialization and vulnerability to climate change for plants in central Europe. This result suggests that highly connected plants are similarly threatened as weakly connected plants. The decline or loss of highly connected plants that interact with many animal partners could have important bottom-up impacts on animal species and ecological networks26,27.

Simulations of species coextinctions indeed demonstrate that mutualistic networks are more sensitive to plant extinction than to animal extinction under climate change in central Europe. This effect could be related to two different mechanisms. First, most studied networks comprised more animal than plant species (Supplementary Data 1) and are thus better buffered against animal than plant extinction11. However, differences between secondary animal and secondary plant extinction were generally larger under climate-induced extinction than under random extinction. As random extinction accounts for differences in animal and plant species numbers and mean specialization in each network, differences in species coextinction between climate change and random extinction must be due to an alternative second mechanism. Different impacts of plant and animal extinction on the networks are, thus, linked to the different relationships between biotic specialization and vulnerability to climate change for plants and animals. As animal species with the highest vulnerability to climate change had a low diversity of plant partners (Fig. 1c,d), the loss of these animal species had a weak impact on the networks and did not disrupt the backbone of interactions in the ecological community (Fig. 2c).

Our results suggest that animal extinction under climate change will only weakly affect animals’ ecological function to plants, such as pollination and seed dispersal. Although this finding has important implications for ecosystem functioning, the simulations did not account for variability in the functional quality of different animal mutualists24. Thus, the inference of our simulation model is limited to quantitative contributions of animals to ecosystem functions18, such as the number of visits by animal pollinators or seed dispersers. Nevertheless, our simulation model suggests a high robustness of plants to animal extinction in future communities. This is consistent with the finding that…