Fragmentation modifies seed trait effects on scatter-hoarders’ foraging decisions

Abstract

Scatter-hoarding animals are crucial in seed dispersal of nut-bearing plants. We used the holm oak Quercus ilex—wood mouse Apodemus sylvaticus mutualism as a model system to evaluate the relative importance of seed size and fat content on scatter-hoarders’ foraging decisions influencing oak dispersal and potential recruitment. We performed a field experiment in which we offered holm oak acorns with contrasting seed size (2 vs 5 g) and fat content (3 vs 11%). Moreover, to test if the strength of these seed trait effects was context-dependent, experimental acorns were placed in small fragments, where natural regeneration is scarce or absent, and forest habitats. In small fragments, rodents had to face increased intraspecific competition for acorns and reduced anti-predator cover during transportation. As a result, they became more selective to ensure rapid acquisition of most valuable food items but, in turn, transported seeds closer to avoid unaffordable predation risks. During harvesting and caching, larger acorns were prioritized and preferentially cached. Fat content only had a minor effect in harvesting preferences. In contrast, in forest sites, where rodent abundance was four times lower and understory cover was well-developed, rodents were not selective but provided enhanced dispersal services to oaks (caching rates were 75% higher). From the plants’ perspective, our results imply that the benefits of producing costly seeds are context-dependent. Seed traits modified harvesting and caching rates only when rodents were forced to forage more efficiently in response to increased intraspecific competition. However, when landscape traits limited cache protection strategies, a more selective foraging behavior by scatter-hoarders did not result in enhanced dispersal services. Overall, our result shows that successful dispersal of acorns depends on how specific traits modulate their value and how landscape properties affect rodents’ ability to safeguard them for later consumption.

Appendices

Appendix 1: Model structure and priors for every response variable

Total foraging time

We modeled the total foraging time invested by mice as a Poisson process that depended on the fragmentation level (forest or fragment) and the number of acorns available. As acorns were in trees located within sites, we introduced tree nested in site as random factors.

$$y _{i,j,k} \sim {\text{Poisson }}(\lambda_{i,j,k} )$$

$$\log (\lambda_{i,j,k} ) = b0_{j,k} + b1 \times {\text{fragmentation}}_{i} + b2_{j,k} \times {\text{number of acorns}}_{i}$$

Prior distributions—parameters at tree level

$$b0_{j,k} \sim {\text{normal}}\quad\left ( {\mu_{k}^{b0} ,\,\sigma^{b0} } \right)$$

$$b1 \sim {\text{normal}} \quad (0, 100)$$

$$b2_{j,k} \sim {\text{normal}}\quad (\mu_{k}^{b2} ,\sigma^{b2} )$$

Hyper-prior distributions—parameters at site level

$$\mu_{k}^{b0} \sim {\text{normal }}\quad(\mu^{s0} ,\sigma^{s0} )$$

$$\mu_{k}^{b2} \sim {\text{normal}}\quad (\mu_{k}^{s2} ,\sigma^{s2} )$$

$$\mu_{\text{k}}^{\text{s2}} = \alpha + \delta \times {\text{fragmentation}}_{k}$$

$\sigma^{b0} \sim {\text{uniform}}\quad (0, 100)$

$$\sigma^{b2} \sim {\text{uniform}}\quad (0, 100)$$

$$\mu^{s0} \sim {\text{normal}}\quad (0, 100)$$

$$\sigma^{s0} \sim {\text{uniform}}\quad (0, 100)$$

$$\sigma^{s2} \sim {\text{uniform}}\quad (0, 100)$$

$$\alpha \sim {\text{normal}}\quad (0, 100)$$

$$\delta \sim {\text{normal }}\quad(0, 100)$$

$y _{i,j,k}$ is the total foraging time of event i at tree j in site k. b0 is the mean time invested in non-fragmented sites, b1 depicts fragmentation effects. b2 represents the effect of the number of acorns available, which depends on forest fragmentation (α and δ). Each focal tree had a mean b0 and b2 effect ($b0_{j,k} , b2_{j,k} )$, centered on the site where the tree is located $\mu_{k}^{b0} ,\,\mu_{k}^{b2}$.

Handling—searching time

We modeled proportion of time invested in handling or searching as a binomial process that depended on the fragmentation level (forest or fragment) and the number of acorns available. As acorns were in trees located within sites, we introduced tree nested in site as random factors.

$$y _{i,j,k} \sim {\text{Binomial }}(p_{i,j,k} , N_{i} )$$

$$\log {\text{it}}(p_{i,j,k} ) = b0_{j,k} + b1 \times {\text{fragmentation}}_{k} + b2_{j,k} \times {\text{number of acorns}}_{\text{i}}$$

Prior distributions—parameters at tree level

$$b0_{j,k} \sim {\text{normal}}\quad (\mu_{k}^{b0} ,\sigma^{b0} )$$

$$b1 \sim {\text{normal}}\quad (0, 100)$$

$$b2_{j,k} \sim {\text{normal}}\quad (\mu_{k}^{b2} ,\sigma^{b2} )$$

Hyper-prior distributions—parameters at site level

$$\mu_{k}^{b0} \sim {\text{normal }}(\mu^{s0} ,\sigma^{s0} )$$

$$\mu_{k}^{b2} \sim {\text{normal}} (\mu_{k}^{s2} ,\sigma^{s2} )$$

$$\mu_{k}^{s2} = \alpha + \delta \times {\text{fragmentation}}_{k}$$

$\sigma^{b0} \sim {\text{uniform}} \quad(0, 100)$

$$\sigma^{b2} \sim {\text{uniform}}\quad (0, 100)$$

$$\mu^{s0} \sim {\text{normal}}\quad (0, 100)$$

$$\sigma^{s0} \sim {\text{uniform}}\quad (0, 100)$$

$$\sigma^{s2} \sim {\text{uniform}}\quad (0, 100)$$

$$\alpha \sim {\text{normal}}\quad (0, 100)$$

$$\delta \sim {\text{normal}}\quad (0, 100)$$

where $y _{i,j,k}$ is either the searching or the handling time of event i at tree j in site k. Here, N is the total time of the foraging ith event. b0 is the mean time invested in non-fragmented sites, b1 depicts fragmentation effects. b2 represents the effect of the number of acorns available, which depends on forest fragmentation (α and δ). Each focal tree had a mean b0 and b2 effect $\left( {b0_{j,k} , b2_{j,k} } \right)$, centered on the site where the tree is located $\mu_{k}^{b0} ,\,\mu_{k}^{b2}$.

Vigilance time

We modeled time invested in vigilance as a zero-inflated binomial process that depended on the fragmentation level (forest or forest fragment) and the number of acorns available. As acorns were in trees located within sites, we introduced tree nested in site as random factors.

$$y _{i,j,k} \sim {\text{Bernoulli }}(p1_{i,j,k} )$$

$$\log {\text{it}}(p1_{i,j,k} ) = a0_{j,k} + a1 \times {\text{fragmentation}}_{k}$$

$$r _{i,j,k} \sim {\text{Binomial}}\quad (p2_{i,j,k} , N_{i} )$$

$$\log {\text{it}}(p2_{i,j,k} ) = \chi_{i} \times \,b1_{j,k} \times {\text{number of acorns}}_{i}$$

Prior distributions—parameters at tree level

$$a0_{j,k} \sim {\text{normal}} \quad(\mu_{k}^{a0} ,\sigma^{a0} )$$

$$a1 \sim {\text{normal}}\quad (0, 100)$$

$$b1_{j,k} \sim {\text{normal}} \quad(\mu_{k}^{b1} ,\sigma^{b1} )$$

Hyper-prior distributions—parameters at site level

$$\mu_{k}^{a0} \sim {\text{normal}}\quad (\mu^{s0} ,\sigma^{s0} )$$

$$\mu_{k}^{b1} \sim {\text{normal }}(\mu_{k}^{s1} ,\sigma^{s1} )$$

$$\mu_{k}^{s1} = \alpha + \delta \times {\text{fragmentation}}_{k}$$

$$\sigma^{a0} \sim {\text{uniform}}\quad (0, 100)$$

$$\sigma^{a1} \sim {\text{uniform}}\quad (0, 100)$$

$$\mu^{s0} \sim {\text{normal}}\quad (0, 100)$$

$$\sigma^{s0} \sim {\text{uniform}}\quad (0, 100)$$

$$\sigma^{s1} \sim {\text{uniform}}\quad (0, 100)$$

$$\alpha \sim {\text{normal}}\quad (0, 100)$$

$$\delta \sim {\text{normal}}\quad (0, 100)$$

$y _{i,j,k}$ is whether the mouse invested time in vigilance or not during event i at tree j in site k, and $r _{i,j,k}$ is the time invested in vigilance during event i at tree j in site k. The probability of investing time in vigilance (p1) depends on fragmentation level (forest or forest fragment). a0 is the mean time invested in non-fragmented sites and a1 depicts fragmentation effects. The proportion of time invested in vigilance (p2) depends on the number of acorns available. b2 represents the effect of the number of acorns available, which depends on forest fragmentation (α and δ). Each focal tree had a mean b0 and b2 effect $\left( {a0_{j,k} , b1_{j,k} } \right)$, centered on the site where the tree is located $\mu_{k}^{bo} ,\,\mu_{k}^{b1}$.

Depletion time

We modeled the depletion time (in seconds) as a Poisson process that depended on the fragmentation level (forest or fragment). As acorns were in trees located within sites, we introduced tree nested in site as random factors.

$$y _{j,k} \sim {\text{Poisson}} \quad(\lambda_{j,k} )$$

$$\log (\lambda_{j,k} ) = \alpha_{i,k} + \beta \times {\text{fragmentation}}_{j}$$

Prior distributions

$$\beta \sim {\text{normal }}(0, 100)$$

Hyper-prior distributions—parameters at site level

$$\alpha_{k} \sim {\text{normal}}\quad (\mu^{\alpha } ,\sigma^{\alpha } )$$

$$\mu^{\alpha } \sim {\text{normal}}\quad (0, 100)$$

$$\sigma^{\alpha } \sim {\text{uniform}}\quad (0, 100)$$

$y _{j,k}$ is the depletion time at tree j in site k. α is the mean time invested in non-fragmented sites and β depicts fragmentation effects. Each site had a mean α _k .

Multiple choice model

We modeled the probability of an acorn being removed taking into account the number and types (fat content and size) of acorns available in each foraging event. This probability depended on acorns fat content and size. As acorns were in trees located within sites, we introduced tree nested in site as random factors.

$$y _{i,j,k} \sim {\text{Multinomial }}\left( {p_{i,j,k} , 1} \right)$$

$$p_{i,j,k} = \frac{{e_{i,j,k} }}{{\mathop \sum \nolimits_{i}^{{N_{i,j,k} }} e_{i,j,k} }}$$

$$\log (e_{i,j,k} ) = b1_{j,k} + b2_{j,k} \times {\text{fat content}}_{i} + b3_{j,k} \times {\text{size}}_{i}$$

Prior distributions—parameters at tree level

$$b1_{j,k} \sim {\text{normal }}\quad\left( {\mu_{k}^{b1} ,\sigma^{b1} } \right)$$

$$b2_{j,k} \sim {\text{normal}}\quad (\mu_{k}^{b2} ,\sigma^{b2} )$$

$$b3_{j,k} \sim {\text{normal}}\quad (\mu_{k}^{b3} ,\sigma^{b3} )$$

Hyper-prior distributions—parameters at site level

$$\mu_{k}^{b1} \sim {\text{normal}}\quad (\mu_{k}^{s1} ,\sigma^{s1} )$$

$$\mu_{k}^{b2} \sim {\text{normal }}\quad(\mu_{k}^{s2} ,\sigma^{s2} )$$

$$\mu_{k}^{b3} \sim {\text{normal}} \quad(\mu_{k}^{s3} ,\sigma^{s3} )$$

$$\mu_{k}^{s1} = \alpha 1 + \delta 1 \times {\text{fragmentation}}_{k}$$

$$\mu_{k}^{s2} = \alpha 2 + \delta 2 \times {\text{fragmentation}}_{k}$$

$$\mu_{k}^{s3} = \alpha 3 + \delta 3 \times {\text{fragmentation}}_{k}$$

$$\sigma^{b1} \sim {\text{uniform}}\quad (0, 100)$$

$$\sigma^{b2} \sim {\text{uniform}} \quad(0, 100)$$

$$\sigma^{b3} \sim {\text{uniform }}(0, 100)$$

$$\sigma^{s1} \sim {\text{uniform}}\quad (0, 100)$$

$$\sigma^{s2} \sim {\text{uniform }}(0, 100)$$

$$\sigma^{s3} \sim {\text{uniform }}(0, 100)$$

$$\alpha 1 \sim {\text{normal }}(0, 100)$$

$$\delta 1 \sim {\text{normal}} \quad(0, 100)$$

$$\alpha 2 \sim {\text{normal }}(0, 100)$$

$$\delta 2 \sim {\text{normal}} \quad(0, 100)$$

$$\alpha 3 \sim {\text{normal}} \quad(0, 100)$$

$$\delta 3 \sim {\text{normal}} \quad(0, 100)$$

$y _{i,j,k}$ is a binary variable that represents whether the acorn i at tree j in site k was removed (1), or not (0). N represents the number of acorns present in that foraging event at that tree. b1 is the probability of being removed given that the acorn is small and has low fat content, b2 is the effect of fat content, and b3 the effect of size. These effects depended on forest fragmentation (α and δ). Each focal tree had a mean b1, b2, and b3 effect ($b1_{j,k} , b2_{j,k} ,b3_{j,k} )$, centered on the site where the tree was located $\left(mu_{k}^{b1} ,\,\mu_{k}^{b2} ,\,\mu_{k}^{b3}\right)$.

First and last acorns removed

We modeled the probability of an acorn being removed first or last as a Bernoulli process that depended on their fat content and size. As acorns were in trees located within sites, we introduced tree nested in site as random factors.

$$y _{i,j,k} \sim {\text{Bernoulli }}(p_{i,j,k} )$$

$$\log {\text{it}}(p_{i,j,k} ) = b1_{j,k} + b2_{j,k} \times {\text{fat content}}_{i} + b3_{j,k} \times {\text{size}}_{i}$$

Prior distributions—parameters at tree level

$$b1_{j,k} \sim {\text{normal }}\left( {\mu_{k}^{b1} ,\sigma^{b1} } \right)$$

$$b2_{j,k} \sim {\text{normal}}\quad (\mu_{k}^{b2} ,\sigma^{b2} )$$

$$b3_{j,k} \sim {\text{normal }}(\mu_{k}^{b3} ,\sigma^{b3} )$$

Hyper-prior distributions—parameters at site level

$$\mu_{k}^{b1} \sim {\text{normal}}\quad (\mu_{k}^{s1} ,\sigma^{s1} )$$

$$\mu_{k}^{b2} \sim {\text{normal}}\quad (\mu_{k}^{s2} ,\sigma^{s2} )$$

$$\mu_{k}^{b3} \sim {\text{normal}}\quad (\mu_{k}^{s3} ,\sigma^{s3} )$$

$$\mu_{k}^{s1} = \alpha 1 + \delta 1 \times {\text{fragmentation}}_{k}$$

$$\mu_{k}^{s2} = \alpha 2 + \delta 2 \times {\text{fragmentation}}_{k}$$

$$\mu_{k}^{s3} = \alpha 3 + \delta 3 \times {\text{fragmentation}}_{k}$$

$$\sigma^{b1} \sim {\text{uniform}}\quad (0, 100)$$

$$\sigma^{b2} \sim {\text{uniform }}\quad(0, 100)$$

$$\sigma^{b3} \sim {\text{uniform}}\quad (0, 100)$$

$$\sigma^{s1} \sim {\text{uniform}}\quad (0, 100)$$

$$\sigma^{s2} \sim {\text{uniform}}\quad (0, 100)$$

$$\sigma^{s3} \sim {\text{uniform}}\quad (0, 100)$$

$$\alpha 1 \sim normal\quad (0, 100)$$

$$\delta 1 \sim {\text{normal}}\quad (0, 100)$$

$$\alpha 2 \sim {\text{normal}}\quad (0, 100)$$

$$\delta 2 \sim {\text{normal}} \quad(0, 100)$$

$$\alpha 3 \sim {\text{normal}}\quad (0, 100)$$

$$\delta 3 \sim {\text{normal}}\quad (0, 100)$$

$y _{i,j,k}$ is a binary variable that represents whether the acorn i at tree j in site k was removed first/last (1), or not (0). b1 is the probability of being removed first or last given that the acorn is small and has low fat content, b2 the effect of fat content, and b3 the effect of size. These effects depended on forest fragmentation (α and δ). Each focal tree had a mean b1, b2, and b3 effect ($b1_{j,k} , b2_{j,k} ,b3_{j,k} )$, centered on the site where the tree was located $\mu_{k}^{b1} ,\;\mu_{k}^{b2} ,\;\mu_{k}^{b3}$.

Dispersal distance

We modeled how far the mice transported the acorns with a normal distribution. We used a logarithmic transformation in other to meet normality. Dispersal distances depended on acorns fat content and size. As acorns were in trees located within sites, we introduced tree nested in site as random factors.

$$\log (y _{i,j,k} ) \sim {\text{normal }}(\mu_{i,j,k} , \sigma )$$

$$\mu_{i,j,k} = b1_{j,k} + b2_{j,k} \times {\text{fat content}}_{i} + b3_{j,k} \times {\text{size}}_{i}$$

Prior distributions—parameters at tree level

$$b1_{j,k} \sim {\text{normal}}\quad (\mu_{k}^{b1} ,\sigma^{b1} )$$

$$b2_{j,k} \sim {\text{normal}}\quad (\mu_{k}^{b2} ,\sigma^{b2} )$$

$$b3_{j,k} \sim {\text{normal }}(\mu_{k}^{b3} ,\sigma^{b3} )$$

Hyper-prior distributions—parameters at site level

$$\mu_{k}^{b1} \sim {\text{normal}} \quad(\mu_{k}^{s1} ,\sigma^{s1} )$$

$$\mu_{k}^{b2} \sim {\text{normal}}\quad (\mu_{k}^{s2} ,\sigma^{s2} )$$

$$\mu_{k}^{b3} \sim {\text{normal}}\quad (\mu_{k}^{s3} ,\sigma^{s3} )$$

$$\mu_{k}^{s1} = \alpha 1 + \delta 1 \times {\text{fragmentation}}_{k}$$

$$\mu_{k}^{s2} = \alpha 2 + \delta 2 \times {\text{fragmentation}}_{k}$$

$$\mu_{k}^{s3} = \alpha 3 + \delta 3 \times {\text{fragmentation}}_{k}$$

$$\sigma \sim {\text{uniform}}\quad (0, 100)$$

$$\sigma^{b1} \sim {\text{uniform}}\quad (0, 100)$$

$$\sigma^{b2} \sim {\text{uniform}}\quad (0, 100)$$

$$\sigma^{b3} \sim {\text{uniform}}\quad (0, 100)$$

$$\sigma^{s1} \sim {\text{uniform}}\quad (0, 100)$$

$$\sigma^{s2} \sim {\text{uniform}}\quad (0, 100)$$

$$\sigma^{s3} \sim {\text{uniform}}\quad (0, 100)$$

$$\alpha 1 \sim {\text{normal }}(0, 100)$$

$$\delta 1 \sim {\text{normal}}\quad (0, 100)$$

$$\alpha 2 \sim {\text{normal}}\quad (0, 100)$$

$$\delta 2 \sim {\text{normal}}\quad (0, 100)$$

$$\alpha 3 \sim {\text{normal}}\quad (0, 100)$$

$$\delta 3 \sim {\text{normal}} \quad(0, 100)$$

$y _{i,j,k}$ is the distance acorn i at tree j in site k was dispersed. b1 is the mean distance mice transport seeds given they are small and have low fat content, b2 the effect of fat content, and b3 the effect of size. These effects depended on forest fragmentation (α and δ). Each focal tree had a mean b1, b2, and b3 effect ($b1_{j,k} , b2_{j,k} ,b3_{j,k} )$, centered on the site where the tree was located $\mu_{k}^{b1} ,\;\mu_{k}^{b2} ,\;\mu_{k}^{b3}$.

Cached

We modeled if mice consumed or stored the acorns as a Bernoulli process that depended on acorn fat content and size. As acorns were in trees located within sites, we introduced tree nested in site as random factors.

$$y _{i,j,k} \sim {\text{Bernoulli}}\quad (p_{i,j,k} )$$

$$\log {\text{it}}(p_{i,j,k} ) = b1_{j,k} + b2_{j,k} \times {\text{fat content}}_{i} + b3_{j,k} \times {\text{size}}_{i}$$

Prior distributions—parameters at tree level

$$b1_{j,k} \sim {\text{normal }}\quad(\mu_{k}^{b1} ,\sigma^{b1} )$$

$$b2_{j,k} \sim {\text{normal }}\quad(\mu_{k}^{b2} ,\sigma^{b2} )$$

$$b3_{j,k} \sim {\text{normal}} \quad(\mu_{k}^{b3} ,\sigma^{b3} )$$

Hyper-prior distributions—parameters at site level

$$\mu_{k}^{b1} \sim {\text{normal}}\quad (\mu_{k}^{s1} ,\sigma^{s1} )$$

$$\mu_{k}^{b2} \sim {\text{normal}} \quad(\mu_{k}^{s2} ,\sigma^{s2} )$$

$$\mu_{k}^{b3} \sim {\text{normal}}\quad (\mu_{k}^{s3} ,\sigma^{s3} )$$

$$\mu_{k}^{s1} = \alpha 1 + \delta 1 \times {\text{fragmentation}}_{k}$$

$$\mu_{k}^{s2} = \alpha 2 + \delta 2 \times {\text{fragmentation}}_{k}$$

$$\mu_{k}^{s3} = \alpha 3 + \delta 3 \times {\text{fragmentation}}_{k}$$

$$\sigma^{b1} \sim {\text{uniform}}\quad (0, 100)$$

$$\sigma^{b2} \sim {\text{uniform}} \quad(0, 100)$$

$$\sigma^{b3} \sim {\text{uniform}}\quad (0, 100)$$

$$\sigma^{s1} \sim {\text{uniform }}\,(0, 100)$$

$$\sigma^{s2} \sim {\text{uniform}} \quad(0, 100)$$

$$\sigma^{s3} \sim {\text{uniform}}\quad (0, 100)$$

$$\alpha 1 \sim {\text{normal}}\quad (0, 100)$$

$$\delta 1 \sim {\text{normal}} \quad(0, 100)$$

$$\alpha 2 \sim {\text{normal}}\quad (0, 100)$$

$$\delta 2 \sim {\text{normal}} \quad(0, 100)$$

$$\alpha 3 \sim {\text{normal}}\quad (0, 100)$$

$$\delta 3 \sim {\text{normal}}\quad (0, 100)$$

$y _{i,j,k}$ is a binary variable that represents whether the acorn i at tree j in site k was cached (1), or not (0). b1 is the probability of being cached given that the acorn is small and has low fat content, b2 the effect of fat content and b3 the effect of size. These effects depended on forest fragmentation (α and δ). Each focal tree had a mean b1, b2, and b3 effect ($b1_{j,k} , b2_{j,k} ,b3_{j,k} )$, centered on the site where the tree was located $\mu_{k}^{b1} ,\,\mu_{k}^{b2} ,\,\mu_{k}^{b3} \,$.

Appendix 2

See Table 4.

Table 4 Location and area of forest fragments used in the study

Appendix 3 Posterior predictive checks of Bayesian models

Mouse foraging activity

See Figs. 4, 5, 6, and 7.

Fig. 4

Posterior predictive check of models of fragmentation effects on depletion time. Blue dots represent mean values of data; bars represent credible intervals of model predictions

Fig. 5

Posterior predictive check of models of fragmentation effects on total foraging time. Blue dots represent mean values of data; bars represent credible intervals of model predictions

Fig. 6

Posterior predictive check of models of fragmentation effects on proportion of time invested in acorn handling. Blue dots represent mean values of data; bars represent credible intervals of model predictions

Fig. 7

Posterior predictive check of models of fragmentation effects on proportion of time invested in vigilant behaviors. Blue dots represent mean values of data; bars represent credible intervals of model predictions

Mouse foraging decisions

See Figs. 8, 9, 10, and 11.

Fig. 8

Posterior predictive check of models of the probability of an acorn being removed during the first three foraging bouts. Blue dots represent mean values of data; bars represent credible intervals of model predictions

Fig. 9

Posterior predictive check of models of the probability of an acorn being removed during the first three foraging bouts. Blue dots represent mean values of data; bars represent credible intervals of model predictions

Fig. 10

Posterior predictive check of models of acorn transportation distances. Blue dots represent mean values of data; bars represent credible intervals of model predictions

Fig. 11

Posterior predictive check of models of the probability of an acorn being cached. Blue dots represent mean values of data; bars represent credible intervals of model predictions