Current global change scenarios demand knowledge on how anthropogenic impacts affect ecosystem functioning through changes in food web structure. Frugivorous mesocarnivores are a key link in trophic cascades because, while their abundance and behaviour are usually controlled by apex predators, they can provide high-quality seed-dispersal services to plant communities. Thus, the recent rewilding of large carnivores worldwide can trigger cascading effects for plants. We investigated the top-down effects of an apex predator (Iberian lynx Lynx pardinus) on seed-dispersal services mediated by two mesocarnivore species (red fox Vulpes vulpes and stone marten Martes foina) at the plant community level by comparing areas with and without lynx in a Mediterranean mountain range in Southern Spain. We collected scats of mesocarnivores (n = 1575) to assess frugivory and seed dispersal of 15 plant species over two consecutive fruiting seasons and two habitat types (open and forest). Specifically, we assessed the effect of lynx presence on (i) seed occurrence and fleshy-fruit biomass per scat, (ii) the number of scats containing seeds, and (iii) the diversity of dispersed seeds. The quantity and diversity of dispersed seeds drastically decreased under predation risk for both mesocarnivore species. Seed dispersal by stone martens was negatively affected by the presence of lynx, with a marked reduction in the number of scats with seeds (93%) and the diversity of dispersed seeds (46%). Foxes dispersed 68% fewer seeds in open habitats when coexisting with lynx, probably leading to differential contributions to seed-dispersal effectiveness among habitats. Our study reveals a novel trophic cascade from apex predators to plant communities. The behavioural responses of frugivorous mesocarnivores to predation risk and the reduction in the intensity of their faecal deposition pattern are probably related to their lower abundance when co-occurring with apex predators. While rewilding apex predators is a successful conservation tool, attention should be paid to cascading effects across food webs, particularly where frugivore megafauna are missing and mesocarnivores provide unique services to plant communities.

Collection/generation of data: 

We used an experimental design comparing five localities with a year-round territorial presence of reproductive Iberian lynx individuals and five localities without lynxes with similar habitat characteristics. Thus, we were able to compare the seed-dispersal patterns of mesocarnivores among predation-risk and predation-free scenarios. Since mesocarnivores can alter their habitat-use patterns under predation risk, we selected two different types of habitats in each locality: a) mature forest with a dense shrub stratum and b) open vegetation with scarce shrubs and isolated trees. Along 1.5 km transects previously selected in each habitat type, we conducted mesocarnivore scat sampling from October to March 2018–2019 and 2019–2020. Sampling was conducted at 1.5-month intervals, allowing each transect to be replicated four times per fruiting season, for a total of 160 visits (four replicates per transect × two habitats × 10 sites × two seasons). Two mammalian scat experts sampled the transects (~3 m in width) twice on the same day with a 1-meter buffer on both sides of the transect. In addition, we performed DNA barcoding analysis on a subsample of 70 stone marten scats to confirm our visual identification (95% success). We estimated the seed content of the collected faeces to quantify the potential effect of predation risk on quantitative seed dispersal. Two experts visually identified the seeds according to morphological concordance with a reference collection (see González-Varo et al. 2021, for a similar approach). The number of seeds in each scat was quantified by counting them under a lens in a Petri dish. When the number of seeds exceeded 100 (n = 24) within one scat, we divided the seed content into four homogeneous quadrants and extrapolated the number of seeds in one quadrant to the entire sample through multiplication. The fleshy-fruit biomass per scat was estimated to quantify the effect of lynx on the fruit consumption behaviour of mesocarnivores (we assumed mesocarnivores ate the whole fruit). We used the number of quantified seeds per scat and the relationship between the average number of seeds per fruit and the average fleshy-fruit biomass obtained from the frubase database (Jordano 2013) for each consumed plant species. We then calculated two metrics of alpha seed diversity for each locality and fruiting season to determine the differences in the effective number of equally common and dominant species among predation risk scenarios: (1) Hill-Shannon and (2) Hill-Simpson (Chao et al. 2014, Roswell et al. 2021). 

To account for the potential effects of fruit availability on fruit consumption and seed dispersal by mesocarnivores, we estimated the available fruit density during the fruiting peak. At the same time that we carried out scat sampling, we visually estimated the coverage percentage and the individual crop size inside plots of 15 × 15 m every 200 m on both sides of the transect. We estimated crop size using a semi-logarithmic fruit abundance index: 0 = no fruits, 1 = 1–10 fruits, 2 = 11–100 fruits, 3 = 101–1000 fruits, 4 = 1001–10000, 5 > 10000. Fruit estimation was performed only along forest habitat-type transects because, in open habitats, fleshy-fruited plants were rare or absent, except for P. bourgaeana. This tree most often occurred on substrates lacking other vegetation. Therefore, we estimated its crop size by searching for adult individuals in a 100 m buffer across the open habitat transects. Peak fruit density (no. fruits / m²) for each plant species was calculated for each locality and fruiting season as the average crop size of all sampled plots along each transect. Fruit diversity was calculated as the Hill–Shannon/Simpson diversity index of available fruits to describe seed diversity.

Processing the data: 

We fitted generalised and linear mixed models (GLMM and LMM) using the R package lme4 v.1.1-27.1 (Bates et al. 2020) to evaluate the effects of Iberian lynx presence, habitat type, and fruit availability on fruit consumption and seed-dispersal patterns of mesocarnivores. We also fitted generalised linear models (GLM) to assess the effects of lynx presence on seed-dispersal diversity by mesocarnivores. We modelled red fox and stone marten data separately for fruit consumption and seed dispersal and analysed the following response variables: (1) seed occurrence per scat (Bernoulli- distributed variable: 0 = absence, 1 = presence), (2) fleshy-fruit biomass per scat (zero not included), (3) number of scats with seeds (zero included), and (4) seed diversity. Seeds of common fig and rare plant species (grouped in ‘Others’) were not included in the modelling because of their low occurrence in carnivore diets but were included in the diversity analyses. We used binomial errors for seed occurrence, Gaussian errors (identity link) on log-transformed data for fruit biomass, Poisson errors (log link) for the number of scats with seeds, and gamma errors (log link) for seed diversity.

We used lynx presence or absence (L) as a fixed effect. Fruit density (FD; no. fruits/m²) was included as a fixed effect for seed occurrence, fruit biomass, and scats with seeds models. Fruit diversity was previously discarded because it resulted in non-significant Pearson’s correlations with seed diversity to avoid overfitting owing to the reduced sample size (n = 20). We also added habitat type (H) to the number of scats with seeds as a fixed term. Interactions FD × L and H × L were also included in the respective models to account for the differential effects of these variables among the predation risk scenarios. Fruiting season (n = 2) was also added as a fixed effect for all response variables. Locality (n = 10) was added as the random effect in all mixed models, except for the seed diversity models. 

Candidate models contained all the possible combinations with a biological sense of fixed effects and interaction terms explained above, compared to models without our variable of interest, lynx presence/absence and a null model (intercept). Due to the reduced sample size for seed diversity, we performed model selection by comparing models including only the fixed term lynx (L) with null models. We ranked the candidate models according to the Akaike Information Criterion adjusted for small sample sizes (AICc). If no other models were within two AICc units from the top-ranked model (∆AICc = 0), the top model was considered the best but if other models were within two AICc units of the top model (∆AICc ≤ 2.0) these were considered to be a set of models with similar support (Burnham and Anderson 2002). We used the MuMIn package v.1.43.17 (Barton 2013) for model selection. We calculated the marginal and conditional coefficients of determination (pseudo-R2) for the selected models (Nakagawa et al. 2017). We carried out post-hoc simple-slope pairwise comparisons to estimate the model coefficients of the interaction terms using the R package emmeans v.1.6.3 (Russell V. et al. 2021). Model residuals were inspected using the DHARMa package v. 0.4.6 (Hartig and Lohse 2022) and a Moran’s I test indicated no spatial autocorrelation (p-values >0.05). All analyses were conducted in R (version 3.6.1; R Core Team 2019).

References:

• Chao, A., Chiu, C.H., & Jost, L. (2014). Unifying species diversity, phylogenetic diversity, functional diversity, and related similarity and differentiation measures through hill numbers. Annual Review of Ecology, Evolution, and Systematics, 45(September), 297–324. 
• González-Varo, J.P., Rumeu, B., Albrecht, J., Arroyo, J.M., Bueno, R.S., Burgos,T., … Traveset, A. (2021). Limited potential for bird migration to disperse plants to cooler latitudes. Nature, 595(7865), 75–79.
• Jordano, P. (2013). Data from: Angiosperm fleshy fruits and seed dispersers: a comparative analysis of adaptation and constraints in plant-animal interactions, Dryad, Dataset.
  Roswell, M., Dushoff, J., & Winfree, R. (2021). A conceptual guide to measuring species diversity. Oikos, 130, 321–338.
• Barton, K. (2013). Package “MuMIn: Multi-model inference” for R. R Package Version 1.9.5, 45.
• Bates, D., Mächler, M., Bolker, B.M., & Walker, S.C. (2020). lme4: linear mixed-effects models. R Package Version 1.1.21. Retrieved from https://github.com/lme4/lme4/
• Burnham, K.P., & Anderson, D.R. (2002). Model Selection and Multimodel Inference: A Practical Information-Theoretic Approach (2nd ed). In Ecological Modelling (Vol. 172). 
• Hartig, F. & Lohse, L. (2022). Package ‘DHARMa’. R Package Version 0.4.6.
• Nakagawa, S., Johnson, P.C.D., & Schielzeth, H. (2017). The coefficient of determination R2 and intra-class correlation coefficient from generalized linear mixed-effects models revisited and expanded. Journal of the Royal Society Interface, 14(134). 
• Russell V.L., Buerkner, P., Herve, M., Love, J., & Singmann, H. (2021). Package ‘ emmeans .’ R Package Version 1.5.4.