Logging and forest edges reduce redundancy in plant–frugivore networks in an old-growth European forest

Authors


Correspondence author. E-mail: joerg.albrecht@staff.uni-marburg.de

Summary

  1. Seed dispersal by frugivores is the basis for regeneration of fleshy-fruited plants in forest ecosystems. Previous studies have reported a decrease in forest specialist frugivores due to logging and forest edges. Forest generalists appear less sensitive and may even increase at forest edges. Such changes in the abundance of frugivores may have consequences for consumer/resource ratios and competition in plant–frugivore networks.
  2. Optimal foraging theory predicts an increase in dietary specialization of animals at low consumer/resource ratios due to reduced competition. A decrease in forest specialists in logged forests should cause decreased consumer/resource ratios, increased dietary specialization and reduced redundancy, whereas an increased abundance of forest generalists at edges may compensate for a loss of specialists.
  3. In Europe's last old-growth lowland forest (Białowieża, Eastern Poland), we recorded fruit removal by frugivores from fleshy-fruited plant species in the interior and at edges of logged and old-growth forests for 2 consecutive years.
  4. The abundance of forest generalists increased at forest edges, whereas specialists were unaffected. Conversely, logging resulted in a decrease in abundance of forest specialists but had no effect on the abundance of generalists. Accordingly, consumer/resource ratios increased from interior to edges and were reduced in the interior of logged forests compared with the interior of old-growth forests. As predicted by optimal foraging theory, a decrease in consumer/resource ratios coincided with increased dietary specialization and a loss of redundancy in the interior of logged forests. Despite low dietary specialization, redundancy was reduced at forest edges as forest generalists dominated plant–frugivore interactions.
  5. Synthesis. We show that a shift in frugivore assemblages at forest edges and increased dietary specialization of frugivores in the interior of logged forests involved a loss of redundancy compared with continuous old-growth forests. This suggests that seed dispersal services in secondary forest habitats depend on an impoverished subset of dispersal vectors and may suffer reduced adaptive potential to changing environmental conditions. Thus, our study highlights the value of old-growth forests for the conservation of frugivore-mediated seed dispersal processes.

Introduction

The mutualism between fleshy-fruited plants and frugivores is an important process in forest ecosystems (Howe & Smallwood 1982). Fleshy-fruited plants depend on seed dispersal by animals to escape from increased seedling mortality near mother plants, to reach adequate microhabitats for regeneration and to ensure gene flow among populations (Janzen 1970; Nathan & Muller-Landau 2000; Schupp, Jordano & María Gómez 2010).

Old-growth forests comprise only 0.2% of all extant European forests (Hannah, Carr & Lankerani 1995; Bengtsson et al. 2000). Recent work has shown that frugivores, particularly forest specialists, are threatened by the conversion of old-growth forest ecosystems into secondary habitats (Newbold et al. 2012). Studies from both temperate and tropical forests have reported that logging and fragmentation can result in a decrease in forest specialist frugivores and in reduced fruit removal (Moran et al. 2004; Kirika, Farwig & Böhning-Gaese 2008; Albrecht, Neuschulz & Farwig 2012). On the other hand, a loss of habitat specialists may be compensated or even overcompensated by less sensitive habitat generalists that are capable of passing habitat boundaries to exploit fruit resources (Farwig, Böhning-Gaese & Bleher 2006; Breitbach et al. 2010; Neuschulz, Botzat & Farwig 2011). Such compositional changes in the abundance of frugivore assemblages may have consequences for consumer–resource dynamics (Fontaine, Collin & Dajoz 2008), competition for resources and the stability of seed dispersal services of entire plant–frugivore associations. Understanding the mechanisms that influence these dynamics requires a network perspective on plant–frugivore interactions.

In recent years, mutualistic plant–animal interactions have been increasingly analysed using a network approach. Such networks represent the interactions between several plants and animals on the level of species assemblages, incorporating species identity and the frequency of pairwise interactions (Jordano 1987). Conceptually, the stability of an interaction network with a given number of species is expected to increase with the number of links and with the evenness in the strength of these links (MacArthur 1955). This suggests that low specialization, that is, low niche differentiation, may contribute to the stability of food webs and mutualistic networks (MacArthur 1955; James, Pitchford & Plank 2012).

In plant–animal mutualisms, the dietary niche of animals is often closely linked to the function that animals perform within the community (Holland & DeAngelis 2010; Blüthgen & Klein 2011). This is emphasized in the term ‘functional niche’ (Loreau et al. 2001). The extent to which animal species differ in their use of plant resources therefore determines the degree of ‘functional complementarity’ and ‘functional redundancy’ in plant–animal interactions (Blüthgen & Klein 2011). A low level of niche differentiation implies functional redundancy, suggesting a higher temporal stability or persistence of the function if some interactions disappear (MacArthur 1955; James, Pitchford & Plank 2012). Further, a low level of niche differentiation should be favoured in a situation of high interspecific competition due to resource limitation, that is, at high consumer/resource ratios (MacArthur & Pianka 1966). Competition may strongly constrain plant–animal mutualisms (Blüthgen et al. 2007; Benadi et al. 2012), which suggests an equilibrium between the redundancy in biotic processes maintained by animals and individual dietary specialization of animals (MacArthur 1955). Plants profit most from high dietary generalization of frugivores, as the number of dispersal vectors increases, while frugivore individuals have to adapt their foraging behaviour in response to the spatio-temporal availability of fruit resources and the density of competitors.

A recent empirical network study on plant–herbivore and host–parasitoid networks reported that diet breadth of consumers and resource availability determine the sensitivity of species interactions to ecosystem perturbation (Valladares, Cagnolo & Salvo 2012). However, this implies that the dietary specialization of animals is a fixed species attribute. According to optimal foraging theory, the diet breadth of animals is a flexible trait and expected to decrease in response to reduced competition at low consumer/resource ratios (MacArthur & Pianka 1966). In line with this, an experimental study has shown that the specialization of pollinators on plants increases at low consumer/resource ratios (Fontaine, Collin & Dajoz 2008). Fontaine, Collin & Dajoz (2008) predicted that perturbation of ecosystems is likely to alter consumer/resource ratios, which in turn may affect the diet breadth and the functional niche of animal mutualists. In support of this hypothesis, Aizen, Sabatino & Tylianakis (2012) showed for plant–pollinator networks that once the most vulnerable species have become extinct, the remaining common and most generalized species begin to specialize and shift from the core to the periphery of the network. Likewise, altered consumer–resource dynamics in degraded forest habitats may influence the foraging behaviour of frugivores. A decrease in the diet breadth of frugivores at low frugivore densities may reduce redundancy, that is, the number of dispersal vectors. Up to now, optimal foraging has not been considered as a driving force of changes in redundancy in plant–animal mutualisms following ecosystem perturbation.

Here, we present a study on the effects of logging and anthropogenic forest edges on consumer/resource ratios, dietary specialization and redundancy in plant–frugivore networks in an old-growth European forest. For 2 consecutive years, we recorded fruit removal by frugivores from fleshy-fruited plants in the interior and at edges of logged and old-growth forests in Europe's best-preserved old-growth lowland forest (Białowieża, Eastern Poland). Based on the results of previous studies, we expected (i) a decrease in the abundance of forest specialist frugivores in logged forests and at forest edges and an increase in the abundance of forest generalists at edges (Farwig, Böhning-Gaese & Bleher 2006; Kirika, Farwig & Böhning-Gaese 2008; Neuschulz, Botzat & Farwig 2011; Menke, Böhning-Gaese & Schleuning 2012). This compositional change in frugivore abundance should result in (ii) reduced consumer/resource ratios, that is, reduced competition, in the interior of logged forests, but a compensation or even overcompensation and increased competition at forest edges. According to optimal foraging theory, we expected that (iii) frugivore specialization on plants increases when competition for resources is reduced (Fontaine, Collin & Dajoz 2008), and that an increase in dietary specialization causes a decrease in redundancy (MacArthur 1955). Thus, we expected a reduction in redundancy in the interior of logged forests, but no change or even an increase in redundancy at forest edges.

Materials and methods

Study area

Our study was conducted in Białowieża Forest, the last European primary old-growth lowland forest, extending over the border between Poland and Belarus. On Polish territory, the forest covers an area of about 625 km2. Within the Białowieża National Park (c. 100 km2), an area of about 60 km2 is strictly protected, and an area of about 45 km2 has potentially never been commercially logged (Falinski 1986; Sokolowski 2004). More than 80% of the remaining forest has been shaped by commercial logging since the First World War (Bobiec et al. 2000; Bobiec 2002). Moreover, most riverine areas of the forest had been cleared during the 16th and 17th centuries for the purpose of hay production, which led to the creation of numerous forest–grassland transitions along rivers (Sokolowski 2004). The core of the Białowieża National Park is an exceptional and rare reference site for studying the impact of anthropogenic habitat degradation on ecological processes in temperate forest ecosystems (Falinski 1986; Bobiec et al. 2000; Bobiec 2002; Tomiałojć & Wesołowski 2004; Niklasson et al. 2010).

Study design

In the study region, the majority of fleshy-fruited plants are primarily associated with ash–alder flood plain forests (Fraxinus excelsior and Alnus glutinosa, Fraxino-Alnetum community; Matuszkiewicz 2001). Thus, we established our study sites within these flood plain forests. We used maps on the distribution of ash–alder forests for the selection of study sites (Falinski 1994) and verified the suitability in the field. The two-factorial design of our study included a total of 10 study sites. We established our study sites in the interior (= 3) and at edges (= 2) of logged forests outside the National Park (stand age: c. 50 years) and in the interior (= 2) and at edges (= 3) of old-growth forests within the National Park (stand age: c. 100–150 years). We refer to forest edges as transitional zones between closed forest and historically managed riverine meadows. The pairwise distance between study sites ranged from 1.3 to 17.9 km (9.1 ± 5.3 km, mean ± SD throughout).

Frugivore observations

We conducted field sampling in all study sites in 2011 and in 2012, due to logistical constraints, in eight of these study sites (for details see Table S1 in Appendix S1 in Supporting Information). From July to October 2011 and 2012, we weekly monitored fruit ripening of fleshy-fruited plants in the study sites and searched for plant species bearing ripe fruits in a radius of 500 m around the centre of each study site. According to availability of fruiting plants, we selected three individuals (= 72), two individuals (= 15) or one individual (= 14) per species for the frugivore observations in each site and year. The number of fruit producing plant species per study site was lower in 2011 (4.3 ± 1.9) than in 2012 (7.3 ± 1.4; Table S1), but did not differ between forest interior and edges or logged and old-growth forests (Table S2). To document frugivore visits on plants, we observed each plant species in each study site and year three times for a period of 6 h starting from sunrise (18 h × plant species−1 × study site−1 × year−1).

We observed plant–frugivore interactions equipped with binoculars from camouflage tents simultaneously at different study sites (7 observers and 14 observers in 2011 and 2012, respectively). We recorded all frugivore species visiting the individual plants, as well as the number of frugivore individuals, the duration of frugivore visits and fruit-handling behaviour. We distinguished between swallowing, crushing, pecking and dropping of fruits. If a group of conspecific frugivores visited a plant and individual behaviour could not be observed simultaneously, we focussed on the individual being most visible. If the behaviour of different species could not be observed simultaneously, we focussed on the rarer species. We were able to observe fruit handling in 78% of all frugivore visits. Of these, we observed swallowing of fruits in 92%, crushing in 4%, pecking in 4% and dropping of fruits in 3% of visits. Proportions do not add to 100% as single visitors handled fruits in various ways: some fruits were swallowed, crushed or pecked, while others were dropped during the same visit. As fruit handling could only be observed in 78% of visits, we used the data on fruit handling in a first step to determine which frugivore species act as seed dispersers on each plant species (i.e. frugivores swallowing, crushing or pecking on fruits of the respective plant species). In a second step, we defined interaction frequency as the number of visits of the identified seed dispersers on a plant species, independent of their fruit handling (visits hereafter). We classified frugivores into forest specialists and generalists (Table S3; Jędrzejewska & Jędrzejewski 1998a,b; Svensson, Mullarney & Zetterström 2009). Forest specialists reproduce exclusively in forest habitats, whereas forest generalists also reproduce in non-forest habitats.

We estimated the crop size of the observed plant individuals by counting their fruits on the day of observation. In the case of trees, we counted the fruits at representative parts of the tree crown and then extrapolated over the whole tree crown. We estimated the crop size three times for each focal plant species, that is, during each of the three observation sessions, and calculated the mean crop size for each plant species in each study site and year. Then, we calculated the total fruit crop (fruit abundance hereafter) within each network by summing the crop size of each plant species in the respective networks. In 2011, we additionally monitored fruit abundance in the study sites along 250-m transects (for details of the sampling protocol see Appendix S1). Fruit abundance along transects correlated positively with fruit abundance in the networks in 2011 (r = 0.91, t = 6.49, d.f. = 8, < 0.001). Therefore, we used the fruit abundance based on total fruit crop in the networks for all further analyses.

Network analysis

For each study site, we constructed two quantitative interaction matrices (for each year separately) based on the frequency of interactions between plants and frugivores except for the two study sites that were sampled in 2011 only (Table S1). In 2011, three focal plants with very low crop size received no visits and were excluded from the network analysis. The total frequency of a frugivore species was defined as the number of visits on all plant species within a network, whereas the visitation rate from the plants' perspective was given by the total number of frugivore visits on a plant species. Thus, we used the marginal totals of the interaction matrices for calculation of the total interaction frequencies (Blüthgen et al. 2007). Standardization of our study design to the same sampling effort per plant species in each study site allowed us to quantify network structure from the plants' perspective. Hence, our study design allows for conclusions about potential consequences for frugivore-mediated seed dispersal processes.

To test our hypotheses, we used a combination of four measures: (i) the consumer/resource ratio, (ii) the specialization of frugivores on plants, (iii) the evenness in the contribution of frugivores to interaction frequencies per plant species and (iv) redundancy, the effective number of dispersal vectors per plant species. To estimate the consumer/resource ratio CRq in the networks, we first divided the number of frugivore visits on each plant species i in a given network by the crop size of the respective plant species as:

display math(eqn 1)

where Ai is the sum of interactions of plant species i, and Fi is the number of fruits of plant species i. To summarize the consumer/resource ratio CRq for each network q, we calculated the mean consumer/resource ratio per plant species weighted by interaction strength of plants as:

display math(eqn 2)

where Ai is sum of interactions of plant species i, and m is the sum of interactions in the network. Consumer/resource ratios were ln(x)-transformed before calculation of the mean for each network q, because consumer/resource ratios showed a strongly skewed distribution (Fig. S1).

To quantify the degree of complementary specialization among frugivores within each network, we compared the observed frequency distribution of interactions with an expected probability distribution that assumes that all species interact with their partners in proportion to their observed frequency totals (Blüthgen, Menzel & Blüthgen 2006; Blüthgen et al. 2007). We calculated the deviation from the expected probability distribution as the standardized Kullback–Leibler distance d′ for each frugivore species j (Blüthgen, Menzel & Blüthgen 2006). Then, we estimated for each network the mean specialization of frugivores ‹inline image› where each frugivore was weighted by its total interactions in the respective network. The index d′ ranges from 0 to 1, where 0 indicates highest possible generalization and 1 indicates highest possible specialization of frugivores on plants. By definition, d′ is a conservative index of specialization, since it is relatively insensitive to asymmetric specialization, which may occur if a frugivore species specializes on a commonly used resource (Blüthgen 2010). Thus, d′ considers not only the diversity of plants used by frugivores but also whether plant resources are used by other frugivores and quantifies the degree of exclusiveness in the resource niches of frugivores (i.e. resource partitioning).

We quantified redundancy Sq in the networks based on Shannon entropy. In contrast to the niche property ‹inline image› of frugivores, our measure of redundancy Sq reflects the plants' perspective and is based on relative interaction frequencies. Since Sq is based on Shannon entropy, it can be partitioned into independent evenness and richness components in a multiplicative manner (Appendix S1; Tuomisto 2012), where the evenness component Eq quantifies the equitability of interaction frequencies among frugivores per plant species. Here, we use Eq for inference about the extent to which changes in redundancy are attributable to changes in the relative contribution of frugivores to interaction frequencies. To quantify evenness Eq and redundancy Sq, we first calculated the Shannon entropy Hi for each plant species i as:

display math(eqn 3)

where aij is the number of visits of frugivore species j on plant species i, and Ai is the sum of interactions of plant species i (Blüthgen et al. 2008). The exponential form inline image (Jost 2006) expresses the ‘effective’ number of frugivore species on plant species i, that is, the number of frugivore species if all were equally common. The evenness in the interaction frequencies of frugivore species' on plant species i is given by:

display math(eqn 4)

where Ji is the number of frugivore species on plant species i (for derivation and justification see Hill 1973). To summarize the redundancy for each network q, we calculated redundancy Sq, the mean effective number of dispersal vectors per plant species weighted by interaction strength of plants as:

display math(eqn 5)

where Ai is sum of interactions of plant species i, and m is the sum of interactions in the network. Likewise, evenness Eq, the mean equitability in the contribution of frugivores to interaction frequencies per plant species, was calculated as:

display math(eqn 6)

Data analysis

We first tested for a relationship between habitat specialization of frugivores (forest generalist and specialist, respectively; Table S3) and habitat types (location: edge vs. interior; logging: logged vs. old-growth) using a quantitative fourth-corner analysis (Dray & Legendre 2008). The fourth-corner analysis requires (i) a site × species community matrix containing the abundance of each frugivore species in each study site, (ii) a trait matrix containing the habitat specialization of each frugivore species and (iii) a habitat matrix containing information on the location (interior vs. edge) and logging activities (logged vs. old-growth) of each study site (Dray & Legendre 2008). To construct the site × species matrix, we first calculated the mean abundance of each frugivore species across the plant species in each study site and year (i.e. the mean visitation rate of each frugivore species in each of the 18 networks during 18 h). Then, we calculated the mean abundance of each frugivore species across the 2 study years for each study site (i.e. the mean abundance of each frugivore species across the two networks per study site). The significance of the relationship between habitat specialization and habitat type was tested with a χ² statistic and a permutation test (9999 iterations). We chose permutation model 1 that permutes the abundances for each species independently and tests the null hypothesis that species are randomly distributed among the habitats (Legendre, Galzin & Harmelin-Vivien 1997; Dray & Legendre 2008).

Secondly, we analysed the variation in the dependent variables of (i) consumer/resource ratio CRq, (ii) evenness Eq, (iii) frugivore specialization ‹inline image› and (iv) redundancy Sq with linear mixed effects models. In these analyses, we treated location (edge vs. interior), logging (logged vs. old-growth) and their interaction as fixed factors. As the data were recorded in 2 years at the same sites, we included site as a random grouping factor and year as a conditional random factor on site. According to the specification of the random terms, the fixed factors were tested against the residual variation among sites to avoid pseudoreplication. Since mean visitation rates on plants in the networks increased with fruit abundance (Pearson correlation on ln(x)-transformed variables: r = 0.58, t = 2.85, d.f. = 16, = 0.012), we included fruit abundance as a continuous covariate in the models to account for differences in resource quantity across study sites and years. As fruit abundance and network size were highly correlated (Pearson correlation on ln(x)-transformed variables: r = 0.69, t = 3.76, d.f. = 16, = 0.0017), we did not include network size as an additional predictor in the analyses. However, by including fruit abundance in the analysis, we implicitly consider differences in network size across study sites. As our study design is unbalanced, the effects of location and logging are not orthogonal. To account for this uncertainty, we used type III sums of squares to estimate the effects of the explanatory variables. Fruit abundance was ln(x)-transformed prior to statistical analysis. We also tested for an effect of second-order interactions between year and the main factors, location and logging. However, we found no significant interactions with year (Table S4). As the sample size in our study is low, we report results of the simpler models only (Table 1).

Table 1. Summary of linear mixed effects models (type III SS) testing the effect of fruit abundance, location (forest interior vs. edge), logging (logged vs. old-growth) and location × logging on (a) consumer/resource ratio CRq, (b) frugivore specialization ‹inline image›, (c) evenness Eq and (d) redundancy Sq of the plantfrugivore networks (= 18) quantified in Białowieża Forest, Eastern Poland. Fruit abundance was ln(x)-transformed prior to statistical analysis
Source of varianced.f.num, d.f.den F P
  1. Significant predictors at a level of < 0.05 are given in boldface type. d.f.num and d.f.den give numerator and denominator degrees of freedom, respectively.

(a) Consumer/resource ratio
Fruit abundance1,6 32.0 0.0013
Location1,62.560.16
Logging1,6 8.93 0.024
Year1,61.710.24
Location × logging1,6 10.2 0.019
(b) Frugivore specialization
Fruit abundance1,6 12.9 0.011
Location1,61.810.23
Logging1,6 21.1 0.0037
Year1,61.910.22
Location × logging1,6 16.0 0.0072
(c) Evenness
Fruit abundance1,6 33.7 0.0011
Location1,6 11.5 0.015
Logging1,61.670.24
Year1,60.08450.78
 Location × logging1,63.250.12
(d) Redundancy
Fruit abundance1,6 11.1 0.016
Location1,6 7.53 0.034
Logging1,6 15.7 0.0074
Year1,65.580.056
Location × logging1,6 9.50 0.022

Finally, we conducted an exploratory path analysis to separate direct and indirect effects of consumer/resource ratios on frugivore specialization, evenness and redundancy. Based on our hypothesis, we constructed an a priori path model that included the direct effects of consumer/resource ratio on frugivore specialization, evenness and redundancy as well as its indirect effects on evenness and redundancy via frugivore specialization (Fig. 3). We further included the covariance between evenness and redundancy into the model. As the sample size in our study is low, we used the data from all 18 networks for the path analysis. However, the significance of the path coefficients was assessed using conservative z-tests with adjusted sample size (n = 10, i.e. the number of study sites).

We are aware that our study design is spatially confounded as the distribution of the remaining old-growth stands is limited to a single relict of preserved forest within the Białowieża National Park being surrounded by logged forest. Therefore, we assessed the extent to which the species turnover in the frugivore assemblages among study sites was related to the spatial and environmental components in our study design. To do so, we used a PCNM analysis (Principal Coordinates of Neighbourhood Matrix) combined with a multivariate redundancy analysis (RDA) and partitioned the variance in the species turnover that was explained by environmental and spatial components in the study design (Appendix S1). All analyses were conducted in r version 2.14.0 (R Development Core Team 2011), using the packages bipartite (network analysis; Dormann et al. 2009), ade4 (fourth-corner analysis; Dray & Dufour 2007), nlme (linear mixed models; Pinheiro et al. 2011) and sem (path analysis; Fox et al. 2011).

Results

During 1818 observation hours (774 h in 2011 and 1044 h in 2012), we recorded 4377 visits (1583 visits in 2011 and 2794 visits in 2012) of 32 frugivore species (29 bird and three mammal species) on 13 plant species (Fig. 1, Table S3). Three bird species were the most frequent visitors, that is, Sylvia atricapilla (1763 visits), Turdus merula (851) and Erithacus rubecula (742). Three understorey woody species received the most visits per 18 h, that is, Prunus padus (137 ± 105 visits), Rhamnus cathartica (119 ± 154) and Euonymus europaeus (62 ± 48).

Figure 1.

Plant–frugivore networks in the interior and at edges of logged and old-growth ash–alder forests in Białowieża, Eastern Poland. Lower bars represent fruit abundance of plants and upper bars frugivore visitation rates drawn at different scales. Linkage width represents interaction frequency. As a summary, networks are based on the mean frugivore visitation rates on plants and the mean crop size of plants across years and sites for each habitat type. Labels depict plants and frugivores, respectively. For species names, see Table S3 in Appendix S1.

The abundance of forest specialists did not differ between forest edges and the interior (χ2 = 6.35, = 0.29), while forest generalists were more abundant at forest edges than in the interior (χ2 = 169.4, = 0.0044). Conversely, forest specialists were less abundant in logged forests than in old-growth forests (χ2 = 20.0, = 0.035), whereas generalists were unaffected by logging (χ2 = 99.1, = 0.20). Accordingly, the consumer/resource ratio in the networks increased from the forest interior to forest edges and was reduced in the interior of logged forests compared with the interior of old-growth forests (Table 1, Fig. 2a). Frugivore specialization on plants increased in the interior of logged forests, compared with the interior of old-growth forests and forest edges (Table 1, Fig. 2b). Evenness was higher in the forest interior than at forest edges and did not vary with logging (Table 1, Fig. 2c). Redundancy was higher in the interior of old-growth forests compared with the interior of logged forests and forest edges (Table 1 and Fig. 2d).

Figure 2.

(a) Consumer/resource ratio CRq [visits fruit−1 18 h−1], (b) frugivore specialization ‹inline image›, (c) evenness Eq and (d) redundancy Sq of plant–frugivore networks in the interior and at edges of logged and old-growth ash–alder forests in Białowieża, Eastern Poland. Given are least square means controlling for fruit abundance and corresponding standard errors. For statistical analyses, see Table 1. For details on calculation of network metrics, see Materials and methods.

The path analysis indicated an indirect positive effect of increased consumer/resource ratios on redundancy via decreased frugivore specialization and a weak indirect negative effect via reduced evenness (Fig. 3). The negative effect via reduced evenness partly counteracted the positive effect via reduced frugivore specialization at forest edges (Fig. 2c,d).

Figure 3.

Path model showing the relations between consumer/resource ratio CRq, frugivore specialization ‹inline image›, evenness Eq and redundancy Sq. Standardized path coefficients are given next to path arrows with significances depicted by NS > 0.1, *< 0.05 and **< 0.01. Explained variance for endogenous variables is denoted by r², and double-headed arrows indicate covariance. All arrow widths are proportional to path coefficients. Note that data from all 18 networks were used for the path analysis. The significance of the path coefficients was assessed using z-tests with adjusted sample size (n = 10, i.e. the number of study sites).

Discussion

The increased abundance of forest generalist frugivores at forest edges caused an increase in consumer/resource ratios, while a loss of forest specialist frugivores in the interior of logged forests resulted in reduced consumer/resource ratios, compared with the interior of old-growth forests (Fig. 2a). In accordance with optimal foraging theory, a decrease in consumer/resource ratios went along with increased frugivore specialization and a loss of redundancy (Fig. 3). However, despite low dietary specialization, redundancy was reduced at forest edges as evenness was lower compared with interiors (i.e. few frugivore species dominated interactions, Fig. 2c,d). A shift in the frugivore assemblages at forest edges and increased dietary specialization in the interior of logged forests thus involved a clear loss of redundancy compared with continuous old-growth forests.

Frugivore community composition and consumer/resource ratios

Previous studies have reported a decrease in forest specialists due to logging and forest edges, whereas forest generalists seem less sensitive and may even increase at forest edges (Farwig, Böhning-Gaese & Bleher 2006; Kirika, Farwig & Böhning-Gaese 2008; Neuschulz, Botzat & Farwig 2011; Menke, Böhning-Gaese & Schleuning 2012). In accordance with these studies, changes in the composition of frugivore assemblages were not random, but related to the habitat specialization of frugivores. Forest generalists, but not specialists, were more abundant at forest edges than in forest interiors. Conversely, forest specialists, but not generalists, were more abundant in old-growth forests than in logged forests. Frugivores are known to track the distribution of fruit resources in the landscape (Tellería, Ramirez & Pérez-Tris 2008), and habitat generalists may be particularly attracted to forest edges, due to widely visible fruit resources (Menke, Böhning-Gaese & Schleuning 2012). The higher visibility of fruit resources may thus have caused the strong increase in frugivore densities at forest margins. Yet, the strong increase in forest generalists at forest edges compared with the forest interior resulted in reduced evenness in the frugivore assemblages as forest generalists dominated the assemblage of seed dispersers (Fig. 2c). Apart from the habitat specialization, the dominance of a subset of frugivores at high densities may also derive from differences in the efficiency of frugivores to track fruit resources (Tellería, Ramirez & Pérez-Tris 2008), and from differences in the overall specialization on fruits in relation to other food types (e.g. invertebrates; Carnicer, Abrams & Jordano 2008). The reduced abundance of forest specialists in logged forests is consistent with a study from Białowieża Forest that reported reduced bird abundance in logged forests compared with old-growth stands in the National Park (Jędrzejewska & Jędrzejewski 1998a). The higher abundance of forest specialists in the old-growth parts of the forest is likely a result of the high habitat quality of the old-growth stands featuring multistorey vegetation layers, standing dead wood, snags and uprooted trees, which provide irreplaceable habitat for a variety of bird species (Tomiałojć & Wesołowski 2004). The compositional changes in the frugivore assemblages entailed increased consumer/resource ratios at forest edges and reduced consumer/resource ratios in the interior of logged forests compared with the interior of old-growth forests. Thus, our results support the hypothesis that anthropogenic perturbation of ecosystems can result in a shift of consumer/resource ratios in mutualistic networks (Fontaine, Collin & Dajoz 2008).

Consumer/resource ratio, dietary specialization and redundancy

The path analysis showed that an increase in consumer/resource ratios went along with a decrease in dietary specialization of frugivores, which is in line with optimal foraging theory (MacArthur & Pianka 1966), and with previous experimental results from plant–pollinator systems (Fontaine, Collin & Dajoz 2008). This suggests that frugivores adapted their foraging behaviour in response to local changes in competition for fruit resources (see Appendix S1). In contrast to Fontaine, Collin & Dajoz (2008), we measured the mean dietary specialization on the level of frugivore assemblages but did not consider changes within single species. Hence, we cannot disentangle the extent to which the observed change in mean dietary specialization of frugivores was related to species turnover among sites or to changes in the foraging behaviour of frugivores. However, as the most common frugivores also tended to be most abundant in the local networks (e.g. three birds, S. atricapilla, T. merula and E. rubecula together accounted for 77% of all visits, Fig. 1), compositional changes should be of minor importance compared with species level changes in dietary specialization. Further, although changes in consumer/resource ratios explained a considerable proportion of the variation in dietary specialization of frugivores (34%), our results also suggest that other factors beyond the density of frugivores and fruits influenced individual foraging decisions. For instance, the selection of fruit resources may be influenced by differences in the nutritional quality and the relative abundance of fruiting plants (Herrera 1984). The path analysis also showed a strong negative relationship between redundancy and the specialization of frugivores on plants. This supports the prediction that increased niche overlap among species results in increased redundancy in biotic processes (MacArthur 1955; Loreau et al. 2001). However, the path analysis also suggests a weak negative effect of consumer/resource ratios on redundancy via reduced evenness (i.e. a subset of frugivores dominated interactions; Fig. 3). This effect partly counteracted the positive effect of increased consumer/resource ratios on redundancy via reduced dietary specialization at forest edges (Fig. 2c,d). To conclude, the shift in the dominance structure of frugivore assemblages at forest edges and increased dietary specialization of frugivores in the interior of logged forests coincided with a clear loss of redundancy compared with continuous old-growth forests.

In contrast to our findings, recent studies from tropical forests suggest no effect of logging and even a positive effect of forest edges on generalization and stability of plant–frugivore associations (Schleuning et al. 2011; Menke, Böhning-Gaese & Schleuning 2012). Plant–frugivore associations are generally more diverse and less specialized in tropical than in temperate ecosystems, and a large proportion of birds feed exclusively on fruits in the tropics (Kissling, Böhning-Gaese & Jetz 2009; Schleuning et al. 2012). The differences in the response of temperate and tropical plant–frugivore associations to forest degradation may derive from higher diversity and lower specialization in tropical systems. Our results support the hypothesis that temperate plant–frugivore associations may be more prone to species loss than those in the tropics (Schleuning et al. 2012).

Potential consequences for seed dispersal processes

We found a clear loss of redundancy in plant–frugivore associations in secondary forest habitats compared with continuous old-growth forests. From the plants' perspective, the loss of dispersal vectors both in the interior of logged forests and at forest edges may entail reduced adaptive potential and a higher vulnerability of seed dispersal services to changing environmental conditions compared with continuous old-growth forests. In fact, some uncertainty remains in our conclusions since we lack information on the relative contribution of frugivores to plant recruitment. However, we observed swallowing of fruits in 92% of frugivore visits, and this is likely to result in dispersal of seeds (Herrera et al. 1994). In addition, Vázquez, Morris & Jordano (2005) have shown that the number of visits is more important for seed dispersal rates than the number of fruits dispersed per visit. This suggests visitation rates are an adequate surrogate for seed dispersal services. Certainly, frugivore species differ in their effect on plant recruitment (Schupp, Jordano & María Gómez 2010). However, in this case, a loss of dispersal vectors may be even more severe, as different frugivore species often act complementary, because frugivore species may differ in their use of microhabitats or in their home-range sizes (Jordano et al. 2007; McConkey & Brockelman 2011). An impoverished set of dispersal vectors might thus also have consequences for the spatial variability in patterns of plant recruitment and the genetic structure of plants (Bleher & Böhning-Gaese 2001; Voigt et al. 2009).

Limitations

So far empirical network data based on field surveys have always been constructed from incomplete samples (Aizen, Sabatino & Tylianakis 2012). Despite a major sampling effort in our study, we certainly did not observe all possible interactions. However, here, we deliberately standardized our sampling effort across study sites to make inferences from our comparison valid. Furthermore, plant species richness, total interaction frequencies, variability in interaction frequencies and network size differed neither between forest interior and edges nor between logged and old-growth forest (Table S2). Accordingly, none of the reported patterns can be attributed to any habitat related sampling artefact. Finally, our results were consistent between both study years, though the networks were larger in 2012 than in 2011 (Table S2). This suggests our results are robust and not confounded by undersampled interactions.

Although we used a replicated study design, our study is limited to a single relict of old-growth forest, and the generality of our conclusions is not known. Still, after accounting for the spatial component in the study design, the environmental component significantly influenced the frugivore composition in the networks (Table S5). Thus, we are confident that the observed patterns are not merely a spatial artefact. Given that the Białowieża Forest is the best-preserved example of old-growth lowland forest in Europe (Bengtsson et al. 2000; Bobiec et al. 2000), we believe that our results provide valuable insights into the dynamics of plant–animal mutualistic networks after ecosystem perturbation.

Conclusions

Within the limitations of our study, we provide evidence that compositional changes in frugivore assemblages alter consumer–resource dynamics, the dietary specialization of animal mutualists on plants and redundancy in plant–frugivore interactions. The loss of dispersal vectors both in logged forests and at forest edges may impose consequences for seed dispersal of fleshy-fruited plants, as (i) it is likely to reduce the adaptive potential under changing environmental conditions, and (ii) it might affect the spatial variability in plant recruitment. Future studies should aim at linking changes in the structure of plant–frugivore networks to spatial patterns of plant recruitment. Overall, our findings from this unique temperate plant–frugivore association strongly imply a high level of plant dependence on a small set of frugivores and higher vulnerability of frugivore-mediated seed dispersal processes than in some tropical ecosystems. Our study highlights the value of old-growth forests for the conservation of frugivore-mediated seed dispersal processes.

Acknowledgements

We greatly thank the administration of the Białowieża National Park, the forestry administrations of Białowieża, Hajnówka and Browsk, and Polish authorities (Ministry of Environment, GDOS [Polish General Directorate of Environment Conservation, Warsaw] and RDOS [Regional Directorate of Environment Conservation, Białystok]) for the permissions to work in Białowieża Forest. We are indebted to the following volunteer ornithologists who assisted with frugivore observations and without whom realization of fieldwork would not have been possible: V. Bohle, F. Bunsen, A. Fritzsch, C. Gayer, I. Grass, N. Grön, J. Hagge, J. Hennlein, C. Herche, M. Hoffmann, C. Höfs, L. Jeske, M. Jung, S. Kaack, E. Kubitz, I. Möller and M. Pönichen. We greatly acknowledge valuable comments of three anonymous reviewers that helped to improve earlier versions of this manuscript. N.F. and D.G.B. are supported by the Robert Bosch Foundation. This study is part of the PhD thesis of J.A. at the University of Marburg. The project is funded by the German Federal Foundation for Environment (DBU) in the framework of a PhD scholarship.

Ancillary