Forest edges show contrasting effects on an austral mistletoe due to differences in pollination and seed dispersal

Authors


Correspondence author. E-mail: ainhoamagrach@hotmail.com

Summary

  1. The alteration and transformation of the areas surrounding native forests due to anthropogenic disturbance can result in differences near this newly created boundary, termed ‘edge effects’.
  2. Our study describes the consequences of edge creation over two successive stages in the life cycle of a hemi-parasitic mistletoe (Tristerix corymbosus), which determine its reproductive success, in temperate austral forests.
  3. We assessed how flower production and pollinator visits change with distance to the nearest forest edge and how these changes affect fruit set, fruit and seed size, and fruit removal by seed dispersers.
  4. Edge effects were dominated by fruit removal, which increased with the distance to the edge, height in the canopy and fruit availability. As a result, plant reproduction (in terms of seeds produced and fruits removed, which putatively leads to higher seed dispersal) decreased strongly near to forest edges. In contrast, visitation rates of the main pollinator (the hummingbird) were unaffected by edges, and their strong effects on fruit set (including the alleviation of quality pollen limitation arising in the forest interior) might be mitigating the decrease in bumblebee visitation near to forest edges.
  5. Synthesis. Our study shows clearly how secondary and tertiary responses to forest edges acted in opposite directions (increasing or decreasing plant reproductive performance), highlighting the need to study several successive processes that impact upon plant fitness under disturbance. Preserving relatively large patches of old-growth forest with low perimeter/area ratios would be key to the habitat requirements of the main disperser and pollinator and thus for mistletoe reproductive performance.

Introduction

The alteration and transformation of the areas that surround native forests as a result of anthropogenic disturbance can result in sharp differences in ecosystem composition, structure or function near this newly created boundary, termed ‘edge effects’ (Angelstam 1992; Wiens et al. 1993; Pickett & Cadenasso 1995; Harper et al. 2005). Nearby edges, some features of the forest will suffer changes beyond their natural range of variation, thereby reducing the fragment's effective area (Ries & Sisk 2004). As a result, forest fragments become composed of two types of habitat: those maintaining the original characteristics of the native forest (i.e. core areas) and those influenced by the fragment's boundary (i.e. edges; Murcia 1995).

Several studies and reviews describe the existence of both primary and secondary responses to forest edge creation (Murcia 1995; Harper et al. 2005). Primary responses are direct consequences of edge creation (e.g. damage to trees and microclimatic, physical or biogeochemical changes). Secondary responses include alterations in the growth, mortality or reproduction of forest-dwelling organisms (Harper et al. 2005). These responses often result in patch-level changes in species composition, further affecting higher-order processes such as species interactions (‘tertiary responses’, sensu Harper et al. 2005). Reported tertiary responses generally focus on antagonistic interactions, for example changes in nest predation and brood parasitism in birds (Murcia 1995; Meiner & LoGuidice 2003; Spanhove, Lehouck & Lens 2009), increased herbivory and seed predation rates (Sork 1983; Cadenasso & Pickett 2000; del Val et al. 2007). Although the effect of the creation of anthropogenic edges over mutualistic interactions has been evaluated in other studies (Jules & Rathcke 1999; Galetti, Alves-Costa & Cazetta 2003; Bach & Kelly 2004; Ness 2004; Ness & Morin 2008), few of them address the effect such disturbances have on mutualisms affecting successive plant life stages. This is unfortunate, as the disruption of one of these interactions in or nearby anthropogenic edges (Restrepo, Gómez & Heredia 1999; Nunes Ramos & Mäes Santos 2006) can trigger cascading effects that further spread to other interactions or throughout the whole ecosystem (Thomas 1984). Indeed, successive stages in a plant's reproductive cycle (e.g. flower production, pollination, seed set, seed dispersal, seed predation) represent multiplicative bottlenecks that may have critical consequences for plant dynamics and they can all be affected in their own way by the proximity to anthropogenic edges. Different suites of anthropogenic disturbances have been shown to range from those having similar, additive effects on plant fitness (Garcia & Chacoff 2007) to those showing opposite-signed effects that mitigate or cancel each other (Francis et al. 2012). To evaluate the net effect of anthropogenic edges on forest plants, we must therefore evaluate its effects on the successive ontogenetic stages and their cumulative contribution to the plant's reproductive success.

This study aims to quantify the relative importance of (tertiary) forest-edge effects over two types of plant–animal interactions, pollination and seed dispersal in the hemi-parasitic mistletoe Tristerix corymbosus. Our study addressed the following specific questions: (i) Do plants located near forest edges produce more flowers than those located in the interior of the forest? (ii) Do plants located at different distances from the nearest edge receive different numbers of visits by the two main pollinators? (iii) Does fruit set change in relation to the distance to the nearest edge? If so, is this related to the identity of the pollinator or the origin of the pollen? (iv) Is fruit removal by the main disperser D. gliroides affected by the proximity to forest edges? To this end, we selected a set of plants located at different distances from the forest edge and measured: (i) the number of flowers, numbers of visits by the main pollinators, pollination success and fruit set, (ii) the effect of pollinator exclusions and hand pollination with pollen from donors located at different distances from the focal plant on their fruit set and (iii) the number of fruits removed by the main disperser, as a surrogate of seed dispersal. We then used all these data to estimate the overall reproductive success of each plant, and relate it to their location in terms of distances to the nearest forest edge.

Materials and methods

Study system

The study was carried out in a native forest fragment in Chiloé Island, Chile (41°52′57.68″ S, 73°39′59.90″ W). The island has a temperate-humid climate with strong oceanic influence (annual precipitation: 2124 mm, mean temperatures: 8.7 °C; Salinas 2008) and sustains a mosaic of old- and second-growth forest fragments embedded within a matrix of prairies, Sphagnum peatlands and shrubby areas dominated by Baccharis sp., Berberis darwinii and the early-successional tree Drimys winterii. Forested areas are a mixture of Valdivian and North-Patagonian temperate rain forests dominated by broad-leaved evergreen species covered by vines and epiphytes, with numerous logs and snags, a dense undergrowth of bamboo thickets (Chusquea sp.) and shrub-covered gaps and open areas (Armesto & Figueroa 1987).

Forest loss in the area has led to a steady increase in the length of the edges between forest and non-forest areas (Echeverría et al. 2008). Edges are characterized by a reduced canopy height (Fig. S1 in Supporting information) and an increased openness (e.g. 1.0 ± 0.2% vs. 5.7 ± 1.1% in forest interior vs. forest egde; Chacón & Armesto 2005) of the canopy and, as a result, higher irradiance levels (e.g. decreasing from 700 to 4 μmoles m−2 s−1 from the open matrix to the forest understorey; Figueroa & Lusk 2001). Comparable changes (e.g. increased canopy openness but reduced relative humidity) also take place along the vertical gradient, that is, from the understorey to the canopy (Parra et al. 2009).

Our study system includes three keystone species: (i) the hemi-parasitic mistletoe Tristerix corymbosus, which is the only flowering species during the winter in the study area and represents a key resource for the overwintering populations of the hummingbird Sephanoides sephaniodes (Smith-Ramírez 1993; Smith-Ramírez & Armesto 1994; Aizen & Ezcurra 1998). At Chiloé's forests, it is hosted almost exclusively by the vine Campsidium valdivianum, although also occasionally by the edge-species Rhaphithamnus spinosus. It produces flowers grouped in inflorescences of 4–14 flowers with conspicuous red, tubular corollas and flower longevity of approximately 1 week (Aizen 2003). Flowers are produced all year-round, although their production peaks in February–June (Smith-Ramírez 1993). T. corymbosus is self-compatible, but it depends strongly upon pollinator visitation for full seed set (Aizen 2005), with the hummingbird Sephanoides sephaniodes being described as the main pollinator. Fruits (single-seeded green berries) mature the next spring (from September to January; Aizen 2003), and in temperate forests of Chile and Argentina, they are only able to germinate following passage through the gut of its only disperser, the marsupial D. gliroides (Amico & Aizen 2000; Amico, Rodriguez-Cabal & Aizen 2011). (ii) The hummingbird S. sephaniodes, a keystone pollinator in austral temperate forests responsible for as much as 20% of pollination services in these forests (Armesto, Smith-Ramírez & Sabag 1996; Aizen, Vázquez & Smith-Ramírez 2002). (iii) The marsupial Dromiciops gliroides, the sole disperser of up to 80% of the fleshy-fruited species in austral temperate forests (Amico & Aizen 2000; Aizen 2005; Smith-Ramírez et al. 2005; Amico, Rodríguez-Cabal & Aizen 2009) and the almost exclusive consumer of T. corymbosus fruits (Rodríguez-Cabal, Aizen & Novaro 2007).

Flower production

To quantify forest-edge effects on plant reproductive effort, we selected 40 individual T. corymbosus plants located at different distances from the nearest edge (0–142 m) within a large forest fragment (433.17 ha, Fig. 1). For every plant surveyed, we recorded the height at which it was growing (using a 6-m telescopic ladder and, for larger distances, a laser range finder, Nikon 550 AS, Tokyo, Japan) and counted the total number of flowers produced at the beginning of the flowering season (7–9 February 2009), including budding, open and senescent flowers. All of the individual plants selected for the study were located on Campsidium valdivianum vines, and no individuals growing on the other possible host, Rhaphithamnus spinosus, were used to avoid confounding effects due to host identity.

Figure 1.

Location of the study area. The enlarged map shows a subset of the sampled forest with the location of the individual plants used for the different experiments relative to forest edges.

Pollinator visits

During March and April 2009, and coinciding with the flowering peak of T. coymbosus in the study area, we censused flower visitation in a second set of plants (n = 21) located at different distances from the nearest forest edge (0–156 m, Fig. 1). On our first visit, we recorded the height at which the plant was growing, the number of (budding and open) flowers for each focal individual and the number of other flowering T. corymbosus present in a 20-m radius around it. Each plant was then observed for 20 min, recording for each pollinator visit the identity of the flower visitor, the total time spent feeding in the plant and the number of flowers visited. Observations were carried out from 8:30 AM to 5:30 PM, and the time at which each individual plant was censused was randomly assigned. Observations were taken on 16 different days, from March 15 to April 10, resulting in a total observation time of 59.3 h. Fruit set was estimated from the number of fruits produced by each plant, measured in November 2009, fruiting peak for the species in the study area in the year sampled. Flowering of C. valdivianum in the study area peaks around September–February (Smith-Ramírez 1993); therefore, mistletoe attractiveness to pollinators was not affected by the host's flowering display. While all Bombus dahlbomii visits in our data set were made by queens, similar to that reported by Aizen (2003), we refer to them as ‘B. dahlbomii bumblebees’ throughout the text because we cannot be certain that, should we have made more observations, we would have also observed visits by workers.

Pollination experiments

We carried out controlled-pollination experiments with a third set of plants (n = 23) located at different distances from the nearest edge (from 0 to 156 m, Fig. 1). These experiments aimed at teasing apart two processes that might contribute to or mitigate putative edge effects on pollination. First, we evaluated whether edge effects on plant pollination could be masked by changes in plant autogamy rates (i.e. if plants that receive less xenogamous pollen are able to produce autogamous seeds). Second, we evaluated whether plants receiving pollen from more distant sources (e.g. from pollinators arriving from other forest patches across the anthropogenic matrix) show enhanced pollination success associated with better pollen quality (i.e. whether inbreeding depression between more related, nearby conspecifics decreases fruit set). For this purpose, we selected 5 branches with at least 15 flowers within each plant and assigned them at random to each of 5 pollination treatments. These included (i) an ‘open’ control treatment, allowing free access to all pollinators and four treatments in which flowers were bagged with bridal veil to prevent pollinator visitation; (ii) a ‘spontaneous autogamy’, in which no further manipulation was exerted; (iii) a ‘geitonogamy’ (aided autogamy) treatment, in which flowers were hand-pollinated with pollen from other flowers of the same plant; (iv) a ‘neighbour xenogamy’ treatment, in which flowers were hand-pollinated with pollen from the nearest 2–3 flowering individuals (< 20 m away); and (v) a ‘far xenogamy’, in which hand pollinations were performed using pollen from 2 to 3 plants located outside the focal forest patch (i.e. in a neighbouring patch at a distance c. 600 m). Pollen was collected daily from the different source plants and used on that same day to pollinate focal plants that had open flowers (i.e. time between pollen collection and hand pollination was never longer than 8 h). In all but the ‘open’ and ‘spontaneous autogamy’ treatments, flowers were emasculated to prevent self-pollen from arriving to stigmas. Hand pollinations were performed by gently applying pollen with a small paintbrush over the stigmas of all open flowers in the branch.

At least 15 days after applying each treatment, we collected the styles from all senescent flowers from all plants except three, which were lost (see Table S1 for the number of styles collected per plant and treatment). Styles were stored in Eppendorf tubes containing FAA solution (formaldehyde, acetic acid, ethanol 5 : 5 : 90), treated overnight with 10 mol L−1 NaOH, stained with 0.1% aniline blue in 0.1 mol L−1 K3PO4, mounted on slides and examined under an epifluorescence microscope (Aizen 2003). We counted the number of pollen tubes just below the stigma surface and used it as a surrogate of pollination success (i.e. the number of viable pollen grains that landed in the stigma and were able to germinate, as in Aizen 2003). During the subsequent fruiting season (November 2009), we counted, measured and weighed fresh fruits and seeds produced at each treatment (with a 0.01 mm and 0.01 g precision, respectively).

Pollinator exclusions

We evaluated whether edge effects affected the relative contribution of the different pollinators to the pollination success of T. corymbosus using selective-exclusion treatments applied to a fourth set of individuals (n = 40) located at different distances from the nearest forest edge (0–142 m; Fig. 1), and selected to have at least 3 branches with a minimum of 15 flowers. Due to accessibility constraints when constructing the exclusions, we only used individuals situated ≤ 6 m in the canopy. During the flowering peak (February 2009), we recorded the height of the focal plant, counted the number of (budding and open) flowers and removed any fruit present at each marked branch (to avoid confusions in future fruit counts; Aizen 2003). We then installed the pollinator-exclusion treatments as follows: (i) an ‘open’ treatment, unexcluded, (ii) a ‘hummingbird exclusion’ treatment, in which a wire mesh (with hexagonal, 2.5-cm cells) cage prevented the access of hummingbirds, but allowed entrance to smaller pollinators (including Bombus dahlbomii) and (iii) a ‘pollinator-exclusion’ treatment, in which a bridal veil bag prevented access of all biotic pollinators. In November 2009, we counted the number of fruits produced per treatment and measured fruit and seed weight of fresh fruits (0.01 mm and 0.01 g precision, respectively). By subtracting seed weight from fruit weight, we also obtained an estimate of pulp weight for each fruit.

Fruit removal

Plants used to record pollinator visits were subsequently followed to measure fruit removal rates. During the fruiting season (November–December 2009), the number of fruits per plant was counted at weekly intervals using binoculars and a counter. For each plant, we carried out three consecutive counts that we then averaged to have a more precise measure of fruit number per plant. Previous studies (Aizen 2003) and preliminary observations showed that mature fruits rarely fall naturally until the very end of the fruiting season. Previous data (Rodríguez-Cabal, Aizen & Novaro 2007; J. L. Celis, personal communication) and observations carried out during the study period (e.g. daytime observations of fruit-bearing individuals using binoculars, inspection of Dromiciops and bird droppings nearby marked mistletoe individuals) indicated that mistletoe fruits were consumed exclusively by D. gliroides. Although we cannot rule out the occurrence of rare events of fruit consumption by birds, it is fair to assume that all fruits that disappeared between counts were consumed by the only known seed disperser, the marsupial D. gliroides (Amico, Rodriguez-Cabal & Aizen 2011).

Although fruit removal is not a direct measure of seed dispersal, the fact that seed germination only occurs after passage through D. gliroides gut makes the survival probability of removed fruits much higher than that of non-removed ones (which is virtually zero), and therefore, fruit removal is a close proxy of seed dispersal.

Data analyses

Data were analysed by means of Generalized Linear Models (GLM) and Generalized Linear Mixed Models (GLMM), using the GENMOD and GLIMMIX procedures in sas 9.1 and 9.2, respectively (sas; SAS Institute, Gary, NC, USA, 2002–03, see Table S2 for variables included in each model and error distributions and link functions applied). Due to the low number of data obtained for Bombus dahlbomii and other types of pollinators during the observation of pollinator visits, the number of flowers visited and the duration of pollinator visits were analysed only for the hummingbird. The models fitted for fruit weight included seed weight as an additional covariate, to account for differences in fruit size directly related to changes in seed size. Throughout the results, due to the limited power of our analyses (number of plants sampled < 40 in all cases), we also report on marginally significant effects (< 0.1) and interpret them as allowing neither the rejection nor the acceptance of the null hypothesis.

Models showing significant spatial autocorrelation among residuals were re-analysed including spatial eigenvectors as predictor variables. We first estimated the spatial autocorrelation of the raw residuals by means of Moran's I, using sam v3.1 (Rangel, Diniz-Filho & Bini 2006). If the model's residuals were spatially autocorrelated, we used spatial eigenvector mapping (SEVM) to avoid biases in parameter estimates and significance levels. SEVM performs a principal component analysis of a connectivity matrix created using the spatial coordinates for each plant, extracts eigenvectors that represent the relationships between plants at different scales and uses them as independent variables in fitted GLM models (Borcard & Legendre 2002; Diniz-Filho & Bini 2005; Griffith & Peres-Neto 2006 for details, see Fig. S2 for results).

Finally, we estimated the combined effect of height and distance to the nearest edge on overall reproductive success (i.e. the number of seeds dispersed per plant) by combining the estimates of all the previous analyses. We began by calculating the predicted number of flowers produced by a ‘mean’ plant at the different combinations of height and distance to the edge. We then multiplied this number by the expected fruit set at different heights and distances to the edge (i.e. the proportion of flowers that turn into fruits), to obtain an estimate of fruit production at each distance and height. Lastly, to estimate the overall reproductive success, we multiplied this fruit production by the estimated probability of fruit removal (which depends also on fruit crop, see Table S3), at the different heights and distances to the edge.

Results

Flower production

Distance to the forest edge significantly affected T. corymbosus flower production, although its effect varied with the vertical location (height) of the individual mistletoes (Table S3, significant distance and height interaction effect). Near forest edges, plants growing closer to the canopy (5–6 m) produced more flowers than those growing near the floor (1–2 m; Fig. 2a); but when progressing towards the forest interior, flower production at all heights converged towards intermediate values (i.e. it decreased for ‘high plants’ and increased for ‘low plants’). Indeed, flower production of plants growing at a medium height (3 m) was unaffected by distance to the edge. Overall, average flower production of plants located close to the edges of the forest (< 40 m) was comparable to that of plants located far from them (211.3 vs. 186.1 at < 40 m vs. > 80 m, all heights pooled).

Figure 2.

Summary of significant edge effects affecting the reproductive system of Tristerix corymbosus in an old-growth forest fragment from Chiloe Island. (a) Flower production increased for plants growing at higher locations under the canopy nearby edges, but not in the forest centre. For simplicity, the relationship between flower production and distance to the edge is shown at three heights (1, 3 and 5 m; one data point per individual plant), although the original model was fitted using height and distance as continuous variables. (b) The number of pollinator visits per plant changed with the distance to forest edges, but the relationship varied among pollinator species. Points represent census made at each individual plant, for each pollinator group (different symbols). (c) The duration of pollinator visits decreased with the distance to the edge for hummingbirds. Points represent the duration of each individual visit measured (i.e. there are several points for each individual plant, which share the same x-value), after accounting for the effect of the random factor ‘plant’ in the calculation of the residuals.

Pollinator visits

Hummingbirds accounted for a significantly higher proportion of visits (95.8% of visits) than bumblebees (only 3.6%) or small bees (0.6%, Table S3). The number and the duration of pollinator visits depended upon the distance to the nearest edge; but for the former, this relationship varied among pollinator types. Hummingbirds carried out the same number of visits near forest edges and in core areas, while the number of bumblebee and small-bee visits increased towards the forest core (Fig. 2b). Hummingbirds made longer visits to plants with larger flower displays (Table S3) and near forest edges (Fig. 2c) and contacted more flowers when visiting plants with fewer conspecific neighbours (Table S3).

Fruit set (the proportion of flowers turning into fruits) was not significantly affected by the distance to the forest edge, the vertical location (height) or any other variable included in the model (number of flowers visited by hummingbirds, flower display and number of neighbours; Table S3).

Pollination experiments

Pollination success (number of pollen tubes per stigma) showed a marginally significant trend to increase towards the forest edge (Fig. 3a), reflected in a comparable, significant trend in fruit weight (Fig. 3c), but not in fruit set or seed weight (Table S4). In contrast, neither pollination success nor fruit or seed weight varied significantly among treatments, but fruit set did – being significantly higher in the far xenogamy than in the open-pollination control, where it was in turn higher than in the geitonogamy, spontaneous autogamy and neighbour xenogamy (Fig. 3b). The effect of pollination treatment did not vary with distance to the forest edge for any of the variables measured (non-significant DISTANCE*TREATMENT interaction; Table S4).

Figure 3.

Summary of significant edge effects affecting the reproductive system of Tristerix corymbosus in an old-growth forest fragment from Chiloé Island. (a) Pollen tube creation increased nearby forest edges. (b) Fruit set varied among pollination treatments. O, open; F, far xenogamy; N, near xenogamy; S, spontaneous autogamy; G, geitonogamy. (c) Fruit weight increased nearby forest edges. The exclusion of certain pollinator species affected plant fruit set (d) and fruit weight (e). (f) Fruit removal increased significantly towards the forest interior. For simplicity, the relationship between fruit removal and distance to the edge is shown at three heights (1, 3 and 5 m; one data point per individual plant), although the original model was fitted using height and distance as continuous variables. In (a) and (c), points represent individual measurements (i.e. there are several points for each individual plant, which share the same x-value), after accounting for the effect of the random factor ‘plant’ in the calculation of the residuals.

Pollinator exclusions

Fruit set decreased significantly from open-pollination controls to hummingbird exclusions to all-pollinators exclusions (Fig. 3d) and increased marginally with flower display (Table S5). Differences between pollinator-exclusion treatments were not influenced significantly by distance to the forest edge (non-significant DISTANCE*TREATMENT interaction; Table S5).

Fruit, seed and pulp weight were unaffected by the distance to the forest edge, but varied among pollinator-exclusion treatments (Table S5). After controlling for the significant, positive effect of seed weight (which was in turn unaffected by the treatments), fruit weight was significantly higher in the open control than in both pollinator exclusions, and comparable among the latter (Fig. 3e). Pulp weight showed a comparable, although only marginally significant, trend (Table S5).

Fruit removal

Fruit removal rates increased significantly and strongly (more than 10-fold) from the edge towards the core of the forest patch, and plants growing high in the canopy had more fruits removed than those near the forest floor and a stronger edge effect (Table S3 and Fig. 3f). Fruit removal increased also strongly for plants with larger fruit crops (Table S3).

Overall reproductive success

Our estimates of reproductive success increased with height and distance to the edge, indicating that the total number of fruits removed (hence, putatively, the total number of seeds dispersed) per plant was up to 10-fold higher for plants located in the forest interior and higher in the canopy than for those growing nearby edges and closer to the forest floor (Fig. 4).

Figure 4.

Effect of the distance to the nearest edge and the height at which the plant was growing on its overall reproductive success (number of fruits produced and removed per plants) of Tristerix corymbosus, estimated from the combined results in Figs 2 and 3. For simplicity, the relationship between reproductive success and distance to the edge is shown at three heights (1, 3 and 5 m; one data point per individual plant), although the original estimates use height and distance as continuous variables.

Discussion

Our results illustrate the existence of a complex set of responses to the formation of forest edges during the process of fragmentation, in which secondary and tertiary responses can both influence final reproductive success of a mistletoe. Although we detected edge effects over both plant–pollinator and plant–disperser interactions, it was the latter that dictated the effect of edges on plant reproductive success. First, distance to edges had a moderate influence on flower production (mediated by height) and did not affect visitation by the predominant pollinator, the hummingbird (although the relative importance of bumblebee visits increased towards the forest core). Similarly, plants growing nearby forest edges had a greater number of pollen tubes and larger fruits, but a comparable fruit set. However, fruit removal rate (a likely surrogate of seed dispersal) decreased exponentially nearby edges. Overall, plant reproductive success (in terms of the total number of fruits removed) decreased exponentially towards the forest edge and linearly for plants growing closer to the forest floor (as compared to those higher in the canopy).

Tristerix corymbosus plants responded to edges by producing a greater number of flowers nearby them – particularly for plants growing higher in the canopy, an unexpected result, according to our initial hypotheses. Additionally, the observed increases in pollination success and fruit size (despite comparable rates of pollinator visitation) suggest that T. corymbosus individuals also increase resource supply to flowers and fruits – in agreement with previous work reporting positive effects of enhanced resource supplies nearby edges on flowering (Hay, Kelly & Holdaway 2008), sometimes counterbalanced by increases in drought stress (Haman 2004). The positive effects of edge neighbourhood on pollination success and fruit size are somehow contradictory, since they are only significant for the pollination experiment and independent on the pollination treatments. In our view, these results are best explained by enhanced resource availability (mainly light; Figueroa & Lusk 2001; Chacón & Armesto 2005) for flower and fruit provisioning nearby edges. Flower provisioning (e.g. starch deposited in the stigma, Reale et al. 2009) has been reported to enhance pollen tube germination in a number of species (Alcaraz, Hormaza & Rodrigo 2010; Julian, Herrero & Rodrigo 2010). The subsequent increase in fruit weight could be an indirect consequence of higher pollen tube vigour or a direct consequence of higher provisioning in the fruit development phase. Such response would be consistent with its detection in hand-pollinated treatments (three of five treatments in the pollination experiment) and its independence of the pollen source (autogamous vs. exogamous).

Indeed, in our study area, T. corymbosus is largely limited to forest areas owing to its dependence on a specific host (the understorey vine C. valdivianum); however, it is a broadly distributed species that also occurs in mixed Valdivian forests and Mediterranean scrubland and thickets along the western slopes of the Andes in Chile and in the north-west of Argentinian Patagonia (Amico, Vidal-Russell & Nickrent 2007), indicating a shade-tolerant, rather than a shade-specialist, character. It is, however, worth noting that, being a hemi-parasitic plant, the effects of (micro)-habitat changes could be mediated by the responses of its host plant.

We did not find any significant responses to edges for the dominant pollinator, the hummingbird S. sephaniodes, responsible for most visits to T. corymbosus flowers and whose exclusion results in large decreases in fruit set and size. Hence, significant edge effects on pollinator visitation only affected the relative contribution of bumblebees, which increased exponentially towards the core. Our results therefore suggest that pollination by hummingbirds, which are often plastic and tolerant to disturbances (Stiles 1975; Johns 1985; Linhart et al. 1987; Feinsinger et al. 1988; Thiollay 1992; Stouffer & Bierregaard 1995), show greater mobility and larger foraging areas than bumblebees (Snow & Snow 1972; Feinsinger 1976; Feinsinger et al. 1988; Kraus, Wolf & Moritz 2009 for Bombus terrestris) and cross frequently the intervening matrix between forest fragments (Magrach, Larrinaga & Santamaria 2012), is relatively resilient to edge effects. Moreover, the strong, positive effect of hummingbird pollination on fruit set and quality (fruit and pulp weight) is probably related to their capacity to carry pollen for fairly large distances, including neighbouring forest patches. Indeed, the results of the pollination experiment show that pollination with pollen from distant and close neighbours respectively improves and decreases fruit set, thus demonstrating that plants face quality pollen limitation (Aizen & Harder 2007). Arrival of xenogamous pollen is probably maximal nearby edges, where it will be transported by hummingbirds moving among neighbouring forest patches.

Contrary to the pollination experiment, edge effects on fruit removal resulted in strong differences for the reproductive success of T. corymbosus individuals. Changes in fruit removal agreed well with known responses of Dgliroides, a species that is negatively affected by forest loss and fragmentation (Rodríguez-Cabal, Aizen & Novaro 2007), tends to avoid low heights (Fontúrbel et al. 2009; Larrinaga, Piazzon & Santamaría unpublished data), but responds to the availability of food resources (e.g. consuming more fruits during the austral summer, Amico, Rodríguez-Cabal & Aizen 2009) and whose fruit removal rates have previously seen to increase with crop size (Morales et al. 2012). Indeed, fruit crop resulted in an exponential increase in fruit removal; hence, higher fruit production of plants growing at high sites nearby the forest edge resulted in increased fruit removal there. This effect was, however, marginal compared to the exponential decrease in fruit removal towards the forest edge. The overall effect of forest edges on plant reproductive success (in terms of total fruits removed; Fig. 4) was therefore primarily determined by the habitat preferences of its exclusive seed disperser: the marsupial D. gliroides.

Although fruit removal does not necessarily equal successful seed dispersal (e.g. if the disperser defecates the seed in the ground or on the non-host species), in T. corymbosus the success of dispersed seeds is radically greater than that of undispersed ones – as germination has been seen to occur only in seeds that have passed through the gut of the marsupial D. gliroides (Amico & Aizen 2000). Moreover, both our own observations and those of researchers with extensive experience in our study area indicate that fruit removal by birds probably represents a rare event (indeed, never observed to date, despite active searches for it). However, as shown by previous studies, larger removal rates in plants with greater crops could lead to smaller probabilities of long-distance dispersal (Morales et al. 2012).

In summary, the interplay between secondary effects of edge creation on plants, pollinators and seed dispersers resulted in a complex suite of tertiary effects. These effects were dominated by frugivore responses to habitat characteristics (distance to the edge and height in the canopy) and resource (fruit) availability and resulted in strong decreases in plant reproduction nearby forest edges. In contrast, the high mobility of the main pollinator and its strong effects on plant reproduction mitigated other effects of fragmentation, including decreased visitation by bumblebees and pollen quality limitation for plants receiving pollen from nearby neighbours. Our study shows clearly how secondary and tertiary responses to forest edges acted in opposite directions (increasing or decreasing plant reproductive performance), highlighting the need to study several successive processes that impact plant fitness under disturbance. These conclusions must, however, be taken with due caution because they are based on a single site and season. Temporal and spatial variations in the ecological context, hopefully addressed in future studies, could potentially alter the responses reported here.

From a management perspective, our results show the importance of maintaining adequate habitat conditions for both the keystone pollinators and the keystone disperser. In both cases, preserving relatively large patches of old-growth forest with low perimeter/area ratios would be key to the habitat requirements of the main disperser and pollinator (Fontúrbel et al. 2010; Magrach, Larrinaga & Santamaria 2012) and thus for the mistletoe reproductive performance.

Acknowledgements

Y. Zuñiga and M.A. Rodríguez-Gironés provided valuable assistance in the field. Logistic support by the Senda Darwin Biological station is gratefully acknowledged. Funding by the BBVA Foundation (project DOSEL) and the Basque Country Government (pre-doctoral fellowship for A.M.) made this study possible. A.R.L. received funding from the JAEDOC program of the Spanish Research Council (CSIC). All authors declare no conflicts of interest.

Ancillary