Invasive species management restores a plant–pollinator mutualism in Hawaii

Authors


Correspondence author. E-mail: causehanna@gmail.com

Summary

  1. The management and removal of invasive species may give rise to unanticipated changes in plant–pollinator mutualisms because they can alter the composition and functioning of plant–pollinator interactions in a variety of ways. To utilize a functional approach for invasive species management, we examined the restoration of plant–pollinator mutualisms following the large-scale removal of an invasive nectar thief and arthropod predator, Vespula pensylvanica.
  2. We reduced V. pensylvanica populations in large plots managed over multiple years to examine the response of plant–pollinator mutualisms and the fruit production of a functionally important endemic Hawaiian tree species, Metrosideros polymorpha. To integrate knowledge of the invader's behaviour and the plant's mating system, we determined the efficacy of V. pensylvanica as a pollinator of M. polymorpha and quantified the dependence of M. polymorpha on animal pollination (e.g. level of self-compatibility and pollen limitation).
  3. The reduction of V. pensylvanica in managed sites, when compared to unmanaged sites, resulted in a significant increase in the visitation rates of effective bee pollinators (e.g. introduced Apis mellifera and native Hylaeus spp.) and in the fruit production of M. polymorpha.
  4. Apis mellifera, following the management of V. pensylvanica, appears to be acting as a substitute pollinator for M. polymorpha, replacing extinct or threatened bird and bee species in our study system.
  5. Synthesis and applications. Fruit production of the native M. polymorpha was increased after management of the invasive pollinator predator V. pensylvanica; however, the main pollinators were no longer native but introduced. This research thus demonstrates the diverse impacts of introduced species on ecological function and the ambiguous role they play in restoration. We recommend incorporating ecological function and context into invasive species management as this approach may enable conservation managers to simultaneously minimize the negative and maximize the positive impacts (e.g. taxon substitution) of introduced species. Such novel restoration approaches are needed, especially in highly degraded ecosystems.

Introduction

Species interactions are a core component of biodiversity, and conservation efforts are increasingly recognizing the need to conserve and manage interactions in addition to the traditional focus on managing single species or identifying regions of high species richness or endemism for conservation (Kiers et al. 2010; Potts et al. 2010). An ongoing conservation challenge is to develop and implement management plans that maintain and restore species interactions essential to ecosystem function, particularly plant–pollinator mutualisms (Traveset & Richardson 2006; Stout & Morales 2009; Potts et al. 2010; Menz et al. 2011). To confront this challenge, we examined the restoration of plant–pollinator mutualisms following the large-scale removal of an invasive wasp that is both a floral visitor and arthropod predator.

Plant–pollinator mutualisms represent a critical but threatened ecosystem function that can determine the success of ecological restoration (Tylianakis et al. 2008; Dixon 2009), and introduced species can disrupt these mutualisms in a variety of ways (Traveset & Richardson 2006). Invasive floral visitors can displace effective pollinators, replace extirpated native pollinators or facilitate native plant–pollinator mutualisms (Traveset & Richardson 2006; Stout & Morales 2009). For this reason, the management of invasive species may give rise to unanticipated changes in plant–pollinator mutualisms (e.g. pollen limitation) if the remaining species are unable to properly restructure their interactions after the invasive species is removed (Burkle & Alarcon 2011). The diverse impacts of introduced species on plant–pollinator mutualisms emphasize the importance of utilizing a functional approach for invasive species management (Zavaleta, Hobbs & Mooney 2001) and the need for novel restoration strategies in degraded ecosystems (Hobbs, Higgs & Harris 2009).

Island plant–pollinator mutualisms are critical to the overall functioning of island ecosystems and are severely impacted by invasive species (Cox & Elmqvist 2000; Dupont et al. 2004). Island pollination systems appear vulnerable to invasive species because of their low taxonomic diversity and lack of co-evolution with continental predators and competitors (Traveset & Richardson 2006). In Hawaii, the extensive adaptive radiations of endemic Hawaiian honeycreepers [Drepanididae (Cabanis)] (Banko et al. 2002) and bees [Hylaeus (Fabricius)] (Magnacca 2007) have occurred in the absence of social insects. The historical absence of social insects in Hawaii (Wilson 1996), which elsewhere are numerically and behaviourally dominant, has magnified the impacts of invasive social insects because endemic pollinators have not evolved appropriate competitive and defensive mechanisms (Zimmerman 1970; Wilson & Holway 2010). The extinctions and declines of important Hawaiian pollinator guilds, notably among honeycreepers (Scott et al. 1988; Banko et al. 2002) and Hylaeus (Magnacca 2007, 2011), have made restoring plant–pollinator mutualisms critical to the long-term integrity of Hawaiian ecosystems. To determine the role of invasive social insects in the restoration of Hawaiian pollination systems and allow sufficient time for plant–pollinator interactions to reassemble, we reduced invasive western yellow jacket Vespula pensylvanica (Saussure) populations in 9-ha management plots for two consecutive years (Hanna, Foote & Kremen 2012a) and examined the response of the floral visitation rates and the fruit production of a functionally important Hawaiian tree species, Metrosideros polymorpha (Gaudich).

The competitive and predatory dominance of invasive social insects is well known (Moller 1996), but the results of studies examining their impact on plant reproduction are highly variable and restricted to social bees and ants. Invasive social wasps in the genus Vespula are considered some of the world's most ecologically damaging invaders and are continuing to spread around the globe (Beggs et al. 2011). The potential impacts of the invasive V. pensylvanica on Hawaiian plant–pollinator mutualisms are multi-dimensional. Vespula prey on both native and introduced pollinators including endemic Hylaeus and the introduced honey bee Apis mellifera (Linnaeus; Wilson & Holway 2010). To further subsidize their energy demand, Vespula consume copious amounts of carbohydrates and have been found to exploit (Moller et al. 1991; Hanna 2012b) and aggressively defend (Thomson 1989; Grangier & Lester 2011) critical carbohydrate resources (e.g. floral nectar and honeydew), thus potentially competing with other floral visitors.

We hypothesized that the frequency and diversity of plant–pollinator mutualisms and the fruit production of M. polymorpha would increase after the large-scale suppression of invasive V. pensylvanica in Hawaii. To test this hypothesis, we investigated the invader's contribution to M. polymorpha fruit production, the plant's mating system and the effect of Vespula on plant–pollinator mutualisms. Specifically, we (i) determined the efficacy of V. pensylvanica as a pollinator of M. polymorpha, (ii) quantified the dependence of M. polymorpha on animal pollination (e.g. self-compatibility and pollen limitation) and (iii) compared the visitation rates of native and introduced pollinators, and the fruit production of M. polymorpha in V. pensylvanica-managed plots vs. V. pensylvanica-unmanaged plots before and after treatment.

Materials and methods

Experimental design

We used a before–after, control–impact experimental design (Green 1979) to observe the effects of annual management of invasive V. pensylvanica on the floral visitation rates and fruit production of M. polymorpha in Hawaii Volcanoes National Park in 2009 and 2010.

Study sites

We randomly selected eight 9-ha study sites within seasonal submontane M. polymorpha woodland within Hawaii Volcanoes National Park between c. 700 and 1100 m elevation. In the four managed 9-ha sites, we deployed 0·1% fipronil chicken bait for one 24-h period for two consecutive years and reduced V. pensylvanica populations by 95 ± 1·2% during the 3 months following treatment. Four unmanaged sites served as experimental controls (Hanna, Foote & Kremen 2012a). The V. pensylvanica population suppression at managed sites actually extended beyond the 9-ha study area and encompassed ≥36 ha due to the foraging distance capabilities of V. pensylvanica (Hanna, Foote & Kremen 2012a). We paired the managed and unmanaged sites to control for environmental variables (e.g. precipitation, elevation and vegetation), randomly allocated treatment within pairs and separated all sites by ≥1 km to maintain site independence (95% of wasps travel ≤200 m from the nest when foraging; Edwards 1980).

Study plant

Metrosideros polymorpha, 'ōhi'a lehua, is a functionally important endemic tree species that has facultative interactions with a diverse array of species, provides a critical nectar resource and habitat for a largely endemic biota, and contributes to the overall biomass and productivity of the ecosystem (Carpenter 1976; Raich, Russell & Vitousek 1997; Gruner 2004). Metrosideros polymorpha is found on all the main islands of the Hawaiian archipelago and occurs in a variety of climate and substrate regimes from sea level to 2500 m a.s.l. (Corn 1979). Peak flowering occurs from February to July, but flowers can be found at any time of the year (Ralph & Fancy 1995). Metrosideros polymorpha has open, red inflorescences, which were comprised of 11·82 ± 0·41 flowers in our study sites that attract native and introduced birds and insect visitors (Carpenter 1976; Corn 1979; Lach 2005; Junker et al. 2010). The hermaphroditic flowers are partially self-compatible, and when pollinated, the ovary develops into a green capsule containing thousands of wind-dispersed seeds that reach full size within a month of anthesis (Carpenter 1976).

Flower visitation by a single V. pensylvanica wasp

To quantify the effectiveness of V. pensylvanica as a pollinator, we monitored the fruit set of flowers after a single V. pensylvanica visit to virgin stigmas, which had previously been protected from pollination by placing a fine-mesh bag over inflorescences with advanced flower buds. We removed the bag and exposed the stigmas to V. pensylvanica visitation within 1–3 days of flowering, to ensure the stigmas were receptive (Carpenter 1976). We allowed the flowers (1–5 per inflorescence) to be visited by a single V. pensylvanica and recorded the visitor's behaviour (e.g. pollen collection, nectar collection, and/or stigma touch). Following the visitation event, we marked the flower pedicel with paint, isolated the stigma with a plastic tube for ≥7 days to prevent the stigma from receiving any additional pollination and monitored for fruit success 2–3 months later. To control for the possible self-pollination of flowers from these manipulations, we used an identical methodology on flowers that received no visitation within the same inflorescence.

Metrosideros polymorpha mating system

To examine the self-compatibility and pollen limitation of M. polymorpha, we compared the fruit set of three inflorescence treatments: (i) self-pollination: bagged with fine-mesh nylon bags to prevent cross-pollination and maximize self-pollination (Carpenter 1976), (ii) supplemental cross-pollination: applied pollen from ≥5 pollen donors collected from synchronously blooming plants >1 km away to all the stigmas within the inflorescence, and (iii) open: received natural visits by pollinators. We used the index of self-incompatibility (ISI) to examine the variation of M. polymorpha self-incompatibility across sites and time. The ISI ranges from self-incompatible (0–0·2), partially self-compatible (0·2–1) and self-compatible (1) (Zapata & Arroyo 1978). The ISI provides a quantitative estimate of the frequency of self-fruit compared with that following supplemental cross-pollination (Zapata & Arroyo 1978). We calculated the mean ISI per site, before and after V. pensylvanica treatment, using the following equation: ISI = (mean self-pollinated fruit production/mean supplemental cross-pollinated fruit production).

Pollen limitation occurs when plants produce fewer fruits than they would with unlimited pollen receipt. We used the index of pollen limitation (PLI) to examine the variation of M. polymorpha pollen limitation across sites before and after V. pensylvanica treatment, using the following equation: PLI = [1−(mean open fruit production/mean supplemental cross-pollinated fruit production)]. The PLI ranges from 0, no pollen limitation, to 1, the highest pollen limitation (Larson & Barrett 2000).

Relative bird abundance

Due to infrequent observations of bird visits to M. polymorpha, we measured the impact of V. pensylvanica on passerine bird abundance by performing nine 8-min point counts within study sites once before and twice after (6 and 10 weeks) V. pensylvanica treatment in 2010. Station locations within sites were located 100 m apart, and we conducted point counts simultaneously in the paired managed and unmanaged sites between 6:00 and 10:00 in good weather.

Insect visitation rates

To determine how the relative frequency of specific plant–pollinator mutualisms changes through time in response to V. pensylvanica treatment, we performed timed focal inflorescence observations. We conducted 10-min focal inflorescence observations for 1–5 inflorescences on 5–8 trees per observation round, simultaneously within the paired sites. To account for tree and inflorescence variation, we recorded the tree height and inflorescence abundance and the inflorescence height and flower abundance. We classified each observed floral visitor into one of the six taxonomic groups: V. pensylvanica, A. mellifera, Hylaeus, Formicidae (Latreille), Diptera (Linnaeus) and Lepidoptera (Linnaeus). In 2009, we conducted a single observation round from 7:00 to 15:00 2 weeks before and 6 weeks after V. pensylvanica treatment. In 2010, to examine M. polymorpha visitation in more detail, we conducted four observation rounds per day from 5:30 to 15:00 2 weeks before and twice (6 and 10 weeks) after V. pensylvanica treatment. To determine the relative frequency of visitors, we calculated the mean site visitation rate (visits per min) for all visitors and for V. pensylvanica, A. mellifera and Hylaeus because they represented >85% of the floral visitors and >94% of the pollinators observed.

Metrosideros polymorpha fruit production

Before and after V. pensylvanica treatment, we randomly selected 5–8 M. polymorpha trees with ≥3 inflorescences in the bud stage at all sites. On each study tree, we assigned an inflorescence to one of the three treatments: ‘No Visitors’ – bagging with fine-mesh nylon bags to prevent visitation – ‘Insects’ – caging with 0·3 × 0·5 m cylinders made of 2·5-cm mesh chicken wire to allow only insect visitation – and ‘All Visitors’ – no bagging to allow bird and insect visitation (methods were adopted from Carpenter 1976). At least one complete trio of treatments was established in each tree. To account for tree and inflorescence variation, we recorded the tree height and inflorescence abundance and the inflorescence height and bud abundance. Three months after the inflorescence flowered, we counted the number of swollen capsules and calculated the fruit production (percentage of flowers setting fruit) for each inflorescence treatment within every study tree.

Vespula pensylvanica abundance and M. polymorpha fruit production

To inform future V. pensylvanica management decisions, we collected Vespula abundance data in conjunction with the M. polymorpha fruit production data. To estimate the total number of wasps (resident and non-resident visitors) in the study sites, we deployed 13 Seabright yellow jacket wasp traps (Seabright Laboratories, Emeryville, CA, USA) baited with 1·5 mL of heptyl butyrate (98%; Aldrich, St. Louis, MO, USA), a strong wasp attractant (Davis et al. 1969), for 5 days per site during each monitoring round (See Hanna, Foote & Kremen 2012a).

Data analysis

To analyse variation in mean relative bird abundance and insect visitation per site, we performed repeated-measures ancovas, using V. pensylvanica treatment as the fixed factor, day as the repeated-measures factor and site pair as the covariate. We analysed variation in mean insect visitation rate per site across and within years, with separate analyses for total visitation rate and visitation rates for the selected taxonomic groups.

To examine the restoration of M. polymorpha fruit production resulting from annual V. pensylvanica management, we performed within- and across-year statistical analyses. To analyse variation in mean fruit production per site across years, we performed separate repeated-measures ancovas for each inflorescence treatment (No Visitors, Insects and All Visitors). The fixed factor was V. pensylvanica treatment and day was the repeated-measures factor. We used a hierarchical nested anova to analyse within-year variation in fruit production. Inflorescence fruit production was nested within tree, tree was nested within site, and site was nested within V. pensylvanica treatment. The fixed factors included: V. pensylvanica treatment, inflorescence treatment and time (pre- and post-V. pensylvanica treatment).

Prior to analysis, we used an arcsine square root transformation to normalize the pollen limitation, self-incompatibility and fruit production data; a log + 1 transformation to normalize the visitation rate data; and a ln transformation to normalize the heptyl butyrate trap data. To correct for Type 1 errors, we used Bonferroni corrections for multiple comparisons. We conducted all statistical analyses in systat 11 (Systat Software Inc. 2004).

Results

Flower visitation by a single V. pensylvanica wasp

We collected V. pensylvanica single visitation fruit set data from 117 flowers in 34 inflorescences on 17 trees. Vespula pensylvanica was never observed contacting the stigma or collecting pollen, whereas wasps were observed collecting nectar from 93·6% of the flowers visited. There was no significant difference between the fruit production of flowers visited by a single V. pensylvanica and flowers receiving no visitation (Wilcoxon signed-rank test, Z = 0·414, P = 0·679). We conducted an additional Wilcoxon signed-rank test using only the fruit production of the first flower visited by V. pensylvanica in the sequence (n = 28), because flowers visited subsequently may mostly be receiving V. pensylvanica-facilitated self-pollen, and this might weaken any positive effect of Vespula visits on fruit production. We found no differences between the fruit production of the first flower visited by V. pensylvanica and flowers receiving no visitation (Wilcoxon signed-rank test, Z = 0·194, P = 0·846).

Metrosideros polymorpha mating system

An average M. polymorpha ISI of 0·214 ± 0·013 across sites and time confirms that M. polymorpha is partially self-compatible, although weakly so. The ISI of M. polymorpha was not significantly different within sites before and after V. pensylvanica treatment (V. pensylvanica treatment × time anova, F1,12 = 0, P = 0·997; Fig. 1).

Figure 1.

Mean (±1 SE) index of self-incompatibility (ISI) and pollen limitation (PLI) within the four managed and unmanaged study sites pre- and post-Vespula pensylvanica treatment. Bars representing each index (ISI and PLI) with different letters are significantly different (P < 0·05, post hoc Tukey tests).

The average PLI across sites and time was 0·46 ± 0·05, suggesting that M. polymorpha is pollen limited. The change in M. polymorpha PLI after V. pensylvanica treatment was significantly different between the managed and unmanaged sites (V. pensylvanica treatment × time anova, F1,12 = 24·15, P < 0·001). The M. polymorpha PLI was significantly lower in the managed sites after annual V. pensylvanica treatment than in all other V. pensylvanica treatment × time categories (anova, F3,12 = 25·341, P < 0·001; Fig. 1).

Relative bird abundance

We recorded 1793 bird detections during 216 point counts in the eight study sites. The relative bird abundance increased by an average of 81 ± 43% 6 weeks and 140 ± 52% 10 weeks after Vespula treatment in the managed compared to the unmanaged sites. The variation in relative bird abundance was significantly different through time between the V. pensylvanica treatments (F2,10 = 14·350, P = 0·027).

Insect visitation rates

We observed 5069 visitors on 28 148 flowers in 1860 inflorescences from 593 trees across the eight study sites over the 2 years. Variation in mean total visitation rate was not significantly different between V. pensylvanica treatments across years and within 2010, but was significantly different within 2009, because 6 weeks after V. pensylvanica treatment, the mean total visitation rate was significantly higher in unmanaged sites compared to managed sites (Fig. 2a; see Table S1 in Supporting Information). Visitation rates of V. pensylvanica were reduced in managed sites compared to unmanaged sites by an average of 98·4 ± 0·9% in 2009 and 97·3 ± 2·1% in 2010 following the annual V. pensylvanica treatment (Fig. 2b). Mean visitation rates of A. mellifera and Hylaeus increased by an average of 595·9 ± 150·5% and 162·6 ± 82·7% in 2009, and 1472·1 ± 406·4% and 763·5 ± 260·8% in 2010 following the annual V. pensylvanica treatment in managed sites, whereas visitation rates of these groups remained at or close to zero in unmanaged sites (Fig. 2c,d). Correspondingly, variation in mean visitation rate per site for each of the three taxonomic groups differed significantly between the V. pensylvanica-managed sites and V. pensylvanica-unmanaged sites within and across treatment years (P ≤ 0·016 in all cases), except for Hylaeus in 2009 (Fig. 2a; see Table S1). There were no significant relationships between the tree and inflorescence characteristics and the insect visitation rates.

Figure 2.

Mean (+1 SE) visitation rate for (a) all visitors, (b) Vespula pensylvanica, (c) Apis mellifera and (d) Hylaeus within the four managed (-●-) and unmanaged (-○-) study sites during each sampling month in 2009 and 2010. Note that the x-axis scales are different among the four graphs. Arrows indicate the timing of the annual V. pensylvanica treatment.*P < 0·05 (from the two-sample t-tests).

Metrosideros polymorpha pollination

We collected fruit set data from 32 351 flowers in 1419 inflorescences from 172 trees across the eight study sites over the 2 years. Following V. pensylvanica treatment in the managed sites, fruit production was increased in ‘All Visitors’ inflorescences by an average of 99·4 ± 17·4% in 2009 and 107·3 ± 44·1% in 2010 and in ‘Insects’ inflorescences by 118·9 ± 18·6% in 2009 and 142 ± 37·1% in 2010 compared to the unmanaged sites (Fig. 3). Fruit production in ‘No Visitors’ inflorescences remained unchanged between V. pensylvanica treatments within and across years (Fig. 3). Correspondingly, variation in mean fruit production per site differed significantly between the two V. pensylvanica treatments across years for ‘All Visitors’ (F3,18 = 16·81, P ≤ 0·001) and ‘Insects’ (F3,18 = 14·964, P ≤ 0·001) inflorescences, but not for ‘No Visitors’ (F3,18 = 0·036, P = 0·999) inflorescences. In 2009 (F2,617 = 5·471, P = 0·004) and 2010 (F2,638 = 3·366, P = 0·035), there was a significant three-way interaction between V. pensylvanica treatment, inflorescence treatment and time (pre- vs. post-treatment), due to no significant differences in fruit production between V. pensylvanica treatments before treatment, but a significantly higher fruit production following treatment in managed sites for ‘All Visitors’ and ‘Insects’ inflorescences compared to unmanaged sites (Fig. 3).

Figure 3.

Mean (±1 SE) fruit production (percentage of flowers setting fruit) within the managed and unmanaged sites for each inflorescence treatment (no visitors, insects and all visitors) pre- and post-Vespula pensylvanica treatment in (a) 2009 and (b) 2010. *P < 0·05 (from two-sample t-tests).

Mean fruit production per site was significantly different between inflorescence treatments for all V. pensylvanica treatment × time categories (anova, P ≤ 0·007 in all cases; see Table S2, Supporting information). Mean fruit production per site was significantly higher in all ‘All Visitors’ and ‘Insects’ inflorescences compared to ‘No Visitors’ inflorescences across all categories (Tukey HSD tests, P ≤ 0·009 in all cases), but there were no significant differences between ‘All Visitors’ and ‘Insects’ inflorescences (see Table S3, Supporting information). There were no significant relationships between tree and inflorescence characteristics and fruit production.

Vespula pensylvanica abundance and M. polymorpha fruit production

Simple linear regression analyses revealed significantly negative relationships between heptyl butyrate trap catch of V. pensylvanica and M. polymorpha fruit production for ‘All Visitors’ (y = 1·076–0·108x, F1,54 = 43·829, P < 0·001, math formula = 0·580) and ‘Insects’ (y = 1·036–0·108x, P < 0·001, math formula = 0·589) inflorescences.

Discussion

Ecosystems are rapidly being transformed and novel ecosystems are being created because of species extinctions and introductions (Hobbs et al. 2006). As a consequence, the composition and function of many ecosystems have been altered and continue to change (Seastedt, Hobbs & Suding 2008; Hobbs, Higgs & Harris 2009). However, the maintenance and restoration of key ecosystem functions are still possible (Hobbs, Higgs & Harris 2009). Our study presents a unique example in which the management of an introduced species (V. pensylvanica) that disrupts plant–pollinator mutualisms and decreases fruit production of an endemic tree species enabled a different introduced species (A. mellifera) to facilitate the plant–pollinator mutualism and increase the fruit production of the same endemic tree species. This result emphasizes the importance of utilizing a functional framework when planning and assessing invasive species management (Zavaleta, Hobbs & Mooney 2001) and the need for novel restoration approaches in degraded ecosystems.

Increasingly, conservation managers are unable to restore historical conditions both because of extinctions (local or global) and the inability to remove all introduced species; alternatively, they should attempt to minimize the negative and maximize the positive impacts of invasive species to preserve ecological function (Hobbs, Higgs & Harris 2009; Stout & Morales 2009). The impact of invasive species on plant–pollinator mutualisms and the reproduction of native plants depend on the characteristics of the organisms involved and the ecological context (Burkle, Irwin & Newman 2007). We found that M. polymorpha is weakly self-compatible and strongly pollen limited, likely reflecting both the past evolution and current ecological context. Metrosideros polymorpha flowers were historically visited by native honeycreepers, and the predominately red flower colour, dimensions of the floral parts and copious nectar secretion suggest this species is adapted to bird pollination (Carpenter 1976). Previous studies revealed that flower-visiting birds transmit M. polymorpha pollen on their head feathers (Corn 1979) and are important pollinators (Carpenter 1976). However, presently, most species of honeycreepers are absent at lower elevations (<1000 m a.s.l.) because of disease-transmitting mosquitoes, introduced predators and degraded habitat (Ralph & Fancy 1995; Banko et al. 2002). Native and introduced birds were observed visiting M. polymorpha in our study sites, but native and introduced insects were the most frequent and numerous visitors. We found that insects were responsible for the majority of M. polymorpha fruit production within our study sites, in accordance with a meta-analysis conducted by Vazquez, Morris & Jordano (2005) that found that the visitation rate is the most important factor in determining the contribution of pollinators.

The ability of M. polymorpha to attract a diverse array of visitors makes it more vulnerable to invasive nectar thieves but more flexible to shifts in the local floral pollinator assemblage (Knight et al. 2005). Nectar and pollen are spatially separated by 1–3 cm (Carpenter 1976; Corn 1979); thus, nectar thieving invasive ants (Lach 2005, 2008) and V. pensylvanica (Hanna 2012b) can deplete and defend the nectar resource and reduce visitation rates of effective pollinators, without ever contacting the reproductive organs (Lach 2008; Junker et al. 2010). Conversely, pollen-collecting insects (e.g. A. mellifera and Hylaeus) are likely contributing to both cross-pollination and pollinator-meditated self-pollination because the separation between the anthers and central style prevents wind-induced contact (Corn 1979).

Large-scale removal of V. pensylvanica, an invasive non-pollinating floral visitor, initiated pollinator behavioural changes leading to higher visitation rates, and consequently, M. polymorpha pollen limitation significantly decreased and fruit production significantly increased. A morphological mismatch with flowers of M. polymorpha enables V. pensylvanica to competitively exploit and antagonistically defend the nectar. Additionally, we observed Vpensylvanica hunting directly from M. polymorpha flowers, and a substantial portion of the diet of Hawaiian V. pensylvanica consists of A. mellifera and Hylaeus (Wilson, Mullen & Holway 2009). However, the significant increase in abundance of these effective pollinators within 6 weeks of V. pensylvanica management, on a time frame far shorter than the time required for the populations to biologically increase, suggests the alteration of pollinator behaviour, rather than population recovery (Wilson & Holway 2010). The exclusion of visitors on floral resources occupied by V. pensylvanica has been found to occur across plant species (Wilson & Holway 2010), and behavioural avoidance of flowers by pollinators has been caused by numerous predators (see Romero, Antiqueira & Koricheva 2011 and therein).

We documented an increase in the floral visitation of effective pollinators and fruit production of M. polymorpha following V. pensylvanica management. However, the restoration implications from our study are limited because no control sites currently exist in Hawaii with no previous wasp invasion and because of the relatively short experimental timeframe. The lack of an uninvaded site prohibits us from determining whether pre-invasion plant–pollinator mutualisms have been restored. Additionally, to determine more accurately the demographic consequences of V. pensylvanica management, restoration of M. polymorpha fruit production needs to be examined throughout the entire flowering season and the relative impact of fruit production on long-lived M. polymorpha needs to be compared to other key components of the life cycle that may be more critical in determining populations dynamics (Knight et al. 2005).

Visitation rates of endemic Hylaeus and introduced A. mellifera increased after V. pensylvanica management, but their relative contributions to the corresponding increase in fruit production of M. polymorpha differed. After V. pensylvanica management, A. mellifera represented 57·3 ± 6·2% of the total floral visitors, whereas Hylaeus represented 13·9 ± 2·1% (Fig. 3). Additionally, Junker et al. (2010) found that A. mellifera contacted the stigma more frequently and deposited significantly more pollen per stigma contact than did pollen-collecting Hylaeus. Thus, the increased visitation rates and effective pollination of introduced A. mellifera were likely the main cause of the increased fruit production of M. polymorpha following V. pensylvanica management. Apis mellifera appears to be acting as a substitute pollinator for M. polymorpha by replacing extinct or threatened bird species in our study system, similar to the role of introduced Japanese white-eye Zosterops japonica for the Hawaiian 'ie'ie vine Freycinetia arborea (Cox 1983).

In our study system and other disturbed ecosystems that lack native pollinators, A. mellifera contributes positively to the pollination of native plants (Dick 2001), but their community-wide effects need to be further examined because their impact on native flora and fauna varies depending on the ecological context (Butz Huryn 1997; Hansen, Olesen & Jones 2002). Apis mellifera was intentionally introduced to Hawaii in 1857 (Snelling 2003). Consequently, the original pollinator community, through competition with numerically dominant A. mellifera, may have already undergone displacement and local extinction (Paini 2004). The change from a native bird and bee fauna to an A. mellifera-dominated pollinator fauna in Hawaii may differentially impact plant reproduction depending on the plant's pollinator specialization and mating system (Aguilar et al. 2006). Apis mellifera has a negative impact on the highly specialized and self-incompatible (<1% autogamous fruit production) Sesbania tomentosa (Hopper 2002), whereas it has a positive impact on the highly generalized and partially self-compatible M. polymorpha. The lower functional redundancy of oceanic island pollinator systems makes them more vulnerable to extinction and range reduction in endemic pollinators (Hansen, Olesen & Jones 2002). As a result, the abundance and general foraging strategy of A. mellifera may make it a critical functional substitute for endemic pollinators, although a subset of the flora may not receive any benefit or may be negatively impacted. In Hawaii and other highly disturbed ecosystems, future research needs to examine the potential benefits and risks of A. mellifera both to native pollinators and the pollination of endemic plant species to formulate appropriate management plans aimed at preserving or restoring plant–pollinator mutualisms (Dixon 2009; Stout & Morales 2009; Kaiser-Bunbury, Traveset & Hansen 2010).

Conventional restoration approaches focus on maintaining species diversity and population size, yet the ecological interactions that underlie habitat restoration are often incompletely understood. Given the importance of pollination, restoration projects cannot assume that plant–pollinator interactions re-establish themselves (Forup et al. 2008). The results of this study linked the large-scale management of an ecologically damaging invader to the increased floral visitation rate of an introduced pollinator and fruit production of a functionally important endemic tree species. Consequently, this research demonstrates the diverse impacts of introduced species on ecological function and their ambiguous role in pollination restoration. Specifically this research emphasizes: (i) the importance of incorporating ecological function and context into invasive species management, (ii) the negative impact of invasive nectar thieves on native plant reproduction, (iii) the taxon-substitute role of introduced pollinators and (iv) the need for novel restoration approaches in highly degraded ecosystems.

Acknowledgements

This material is based upon work supported by the National Science Foundation Graduate Research Fellowship under Grant No. (DGE 1106400). Funding was provided by the Pacific Island Ecosystems Research Center and the Natural Resources Protection Program of the US Geological Survey. Any use of trade, product or firm names in this publication is for descriptive purposes only and does not imply endorsement by the US government.

Ancillary