Coinvasional disruptions to island pollinator networks

Abstract

Biological invasions can adversely affect pollinator diversity by threatening the maintenance of animal‐pollinated plant communities. Although most studies have examined single invasive species, accelerating species introductions driven by global trade highlight the need to understand how multiple co‐occurring invaders jointly influence pollination networks. We leveraged the relative simplicity of the Ogasawara archipelago to assess the combined impacts of multiple invaders on pollination networks. We surveyed floral and pollinator communities and quantified plant–pollinator interactions across four islands with varying degrees of invasion. We characterized invasion degree based on the abundance of three invasive taxa: green anoles (Anolis carolinensis), a predator; western honey bees (Apis mellifera), a floral competitor; and several flowering plants. Structural equation models were used to quantify direct and indirect effects of these invaders on effective plant–pollinator partners (niche breadth or variety of partners a species interacts with in a given community) and the degree of specialization (dʹ, niche partitioning or selectivity of a species) of plant and pollinator communities. All invaders strongly affected pollinator diversity and, to a lesser extent, plant diversity. Anoles had the strongest negative effect on pollinator diversity, triggering cascading effects across the trophic chain. Directly or indirectly, all invaders negatively affected plant resource use. Honey bee abundance had a direct negative effect on plant effective partners and specialization, and anole abundance and the proportion of invasive plants indirectly affected plant resource use through a reduction in pollinator diversity. Degree of invasion and network structure mediated whether honey bees contributed to functional compensation or facilitated invasional meltdown. Our findings demonstrated that co‐occurring invaders generated synergistic pressures on pollinators and plants, underscoring the importance of evaluating invasions in a multistressor context. Integrative approaches are essential for developing conservation strategies in vulnerable ecosystems, particularly oceanic islands, where biological invasions are the principal driver of biodiversity loss.

INTRODUCTION

Accelerating global trade has significantly increased species translocation, boosted immigration rates, and altered community composition (Gleditsch et al., 2023). These dynamics have contributed to alarming declines in insect pollinators (Wagner, 2020) and have had substantial ecological and economic consequences (Ollerton, 2017). Pollinator declines are particularly concerning because their interdependence with flowering plants shapes complex interaction networks that form the foundations of biodiversity (Bascompte & Scheffer, 2023). Quantitative field studies are needed to capture the mechanistic processes driving pollinator loss and assess the relative contributions of interacting factors (Weisser et al., 2023).

Invasive non‐native species (hereafter invasive species) (as defined by Soto et al. [2024]) interact with native species either through intratrophic competition or through intertrophic predator–prey dynamics (Gonzales & Arcese, 2005). However, research on biological invasion impacts on ecological networks has focused on intratrophic interactions and individual invasive species (Traveset & Richardson, 2014). This approach overlooks two crucial aspects of invasion biology: ecosystems commonly harbor multiple invasive species (Gurevitch et al., 2011) and species are typically linked across trophic levels (Schuldt et al., 2018). Impacts from invasion can thus cascade through ecological networks, leading to top‐down or bottom‐up effects (Leroux & Loreau, 2015). Furthermore, impacts can percolate into other layers of complexity, with each layer corresponding to an ecological function (Hervías‐Parejo et al., 2024). Invasive species can also lead to a reshuffling of pairwise interactions (interaction rewiring), affecting species persistence and network functioning (Poisot & Stouffer, 2016). To better understand the intricate dynamics of coinvasions and their effects on ecosystem functions, a quantitative assessment of both intra‐ and intertrophic interaction networks is needed.

Islands can act as indicators of environmental change because insular ecosystems are more prone to extinction and more vulnerable to invasive species than those on the mainland. Global trade has altered natural patterns of island isolation from species source pools, enabling species introduction from locations connected economically (Gleditsch et al., 2023). Indeed, human‐induced biological invasions are the primary threat to island ecosystems globally, causing diverse ecological impacts on native flora, fauna, and the interactions between them (Traveset & Richardson, 2014).

Invasive plants often compete for the attention of native pollinators, reducing pollinator visitation rates and reproductive success of native plants (Emer & Timóteo, 2020; Morales & Traveset, 2009). Invasive animals can disrupt plant–pollinator dynamics by acting as novel competitors or predators of native pollinators (Montero‐Castaño & Vilà, 2017; Spiller & Schoener, 2008). The western honey bee (Apis mellifera), for instance, may outcompete native pollinators, which could lessen native pollinator diversity, decrease visitation rates, and eventually lead to low native plant reproductive success (Fumero‐Caban et al., 2022; Pasquali et al., 2025). Invasive predators pose a particular threat on islands because selection for predator defenses in native fauna is limited (Blumstein & Daniel, 2005).

The introduction of invasive predators has triggered substantial declines in biodiversity on various islands, such as Guam (Gurevitch & Padilla, 2004), Taiwan (Huang et al., 2008), Ogasawara (Abe et al., 2008; Sugiura, 2016), and the Bahamas (Spiller & Schoener, 2008). However, invasive species may also have positive effects in island ecosystems (reviewed in Etter et al., 2022; Traveset & Richardson, 2014). Invasive flora can enhance pollination of native plants through “magnet” species effects when increased collective floral display and diversity result in increased floral visitation (Ghazoul, 2006). Invasive fauna can also act as beneficial partners, compensating for the extinction of native pollinators (Traveset & Richardson, 2014). Consequently, island ecosystems are a conservation priority and exceptional natural laboratories for studying the impacts of invasive species.

The oceanic Ogasawara Islands of Japan exhibit unusually high levels of endemism (90% bees, 40% plants) (Kato et al., 1999). Numerous invasive species are also present in varying abundance depending on the history of human occupation. We focused our study on four islands that can broadly be categorized as either heavily invaded (Chichijima and Hahajima) or lightly invaded (Anijima and Ototojima) based on the duration of the invasion and presence and density of invasive species, including plants, a predatory anole, and a bee pollinator (see “METHODS” for details of historical disturbance). Recent declines in native pollinators in the two inhabited islands have been attributed to invasive plant and animal species (Abe et al., 2008) and have had knock‐on effects for the pollination of endemic plants (Sugiura et al., 2006; Watanabe et al., 2017). Although the effects of invaders are often qualitatively evident, identifying key causal factors and understanding the community‐wide consequences of invasions require quantitative assessment at an ecological network level.

We used interaction network analyses to identify the potential effects of invaders through top‐down or bottom‐up mechanisms from plant and pollinator perspectives. We explored, directly and indirectly via changes in plant and pollinator diversity, the combined effects of invasive species on the structure of pollination networks. We investigated how these disruptions might affect different pollinator functional groups and assessed the vulnerability of these groups with a view to guiding future conservation management decisions.

METHODS

Study design

The Ogasawara oceanic archipelago (27°10ʹ to 26°39ʹN; 142°09ʹ to 142°13ʹE) is situated ∼1000 km south of Tokyo, Japan, in the northwestern Pacific Ocean (Figure 1). The area is characterized by a subtropical climate. Monthly mean temperatures and rainfall ranges are 18–28°C and 60–170 mm, respectively (Kawakami & Okochi, 2010). We established sampling sites on four islands (Figure 1), two inhabited (Chichijima and Hahajima) and two uninhabited (Anijima and Ototojima).

FIGURE 1.

Study sites on the Ogasawara archipelago surveyed to quantify the effects of invaders on effective partners and species‐level specialization in plant and pollinator communities across four islands with varying degree of invasion (CHI, Chichijima; HA, Hahajima; CHI_N and HA_N, natural forest; CHI_D and HA_D, disturbed forest; ANI, Anijima; ANI_I, anole‐invaded; ANI_F, anole‐free; OTO, Ototojima; yellow areas indicate forests replanted with native plant species). Details on study design are in Appendix S5.

The four islands sampled differ markedly in their degree of human disturbance and biological invasion. Chichijima and Hahajima are heavily disturbed by forestry and agriculture and consist of a mix of protected reserves and inhabited areas of degraded habitat dominated by invasive plant species (Abe et al., 2008; Toda et al., 2010) (Appendix S1). Anijima and Ototojima are relatively lightly disturbed. Their secondary forests are well conserved, and there are few invasive plants (Abe et al., 2008) (Appendix S1). We used GIS maps (Japan Ministry Environment) to inform site selection and to define two broad vegetation categories: natural forest, which included subtropical evergreen broad‐leaved forest, shrubland, and grassland, and disturbed vegetation, which included secondary forests in urban and agricultural matrices (Chichijima and Hahajima) or native reforestation areas (Anijima and Ototojima).

Apiculture was introduced to the islands in the late 19th century, and apiaries are currently maintained on the inhabited islands (Chichijima 52 hives and Hahajima 15 hives). This, combined with unmanaged feral colonies on some uninhabited islands, has created a gradient of honey bee abundance across the islands; abundance decreases as distance to the apiaries increases (Abe et al., 2008, 2011) (Appendix S2). The archipelago has also been invaded by the green anole (Anolis carolinensis) (hereafter anole), an arboreal insectivorous lizard that uses a diurnal sit‐and‐wait ambush strategy around flowers (Losos, 2011). Their diet consists predominantly of arthropods (up to 90%), including lepidopteran and hymenopteran pollinators and arachnid predators (Huang et al., 2008). Their predation may significantly affect island food webs (Schoener & Spiller, 1999). Anoles became established in the 1960s on Chichijima and spread in the 1980s to Hahajima. Southern Anijima was invaded around 2013, prompting the erection of an anole‐proof fence to limit their spread (Kanto Regional Environment Office, 2022). No anoles have been observed in Ototojima.

Using information from GIS maps and previous studies (Abe et al., 2008; Toda et al., 2010) on variation in the presence and abundance of invasive plant and animal species across the study islands, we established seven pairs of sampling sites representing different combinations of invasive species. We refer to these site pairs as invasional contexts. Each invasional context was defined by a combination of habitat disturbance—whether sites were located in disturbed vegetation areas or in natural forest—and the abundance of invasive honey bees and anoles (Appendix S2). These categories enabled us to group sites based on shared ecological and invasional characteristics for subsequent comparative analyses.

Each invasional context consisted of two 300 × 4‐m sampling sites, for a total of 14 sites (seven paired invasional contexts) for the four islands. On each of Chichijima (CHI) and Hahajima (HA), islands with variation in habitat disturbance but consistently high anole and honey bee densities, we established two sites in natural forests (hereafter natural invasional context) (CHI_N and HA_N) and two sites embedded in disturbed vegetation (hereafter disturbed invasional context) (CHI_D and HA_D). In natural forests, the mean distance to the nearest disturbed vegetation area was 709.68 m (SD 482.29) on Chichijima and 94.92 m (45.00) on Hahajima.

Natural forests had relatively low probabilities of invasive plant presence compared with disturbed vegetation areas, which allowed us to investigate the effects of invasive plants. In Anijima, which has uniform forest conditions (mean distance to secondary forest 499.02 m [SD 140.79]) and low honey bee densities (mean distance to apiaries 2451.95 m [262.83]), we established two sites in the anole‐invaded area (hereafter anole‐invaded invasional context [ANI_I]) and two sites in the anole‐free area (hereafter anole‐free invasional context [ANI_F]). The distance between anole‐invaded and anole‐free sampling sites was approximately 450 m, and these sites were intersected by the anole‐proof fence. This allowed us to independently assess the influence of anole abundance on the pollination network.

Lastly, in Ototojima, which has relatively uniform forest conditions across the island (mean distance to secondary forest 220.87 m [SD 169.93]), very few honey bees (mean distance to apiaries 5994.15 m [116.28]), and no anoles, we established two sites constituting one invasional context (OTO) (Figure 1). Initial site planning, based on GIS and previous data (Appendix S2), was confirmed by our recorded observations (see “RESULTS”).

To capture seasonal variation in plant and pollinator communities, we sampled in May, July, and September, which covered the full flowering peak in 2021. We sampled each site once per sampling period, resulting in three temporal replicates for all recorded variables at each site.

Diversity measures

To characterize the plant and pollinator community, we recorded abundance and species richness of all flowering plants and censused pollinators on flowers during every sampling period at each site. We used the field guide Flora of the Bonin Islands (Toyoda, 2003) to classify flowering plants as native or invasive. Open flower abundance was estimated per plant individual by counting flowers on five randomly selected inflorescences across five randomly selected branches. We extrapolated to the whole individual and then summed across individuals to quantify species‐ and site‐level flower availability. We calculated the plant invasion level as the proportion of invasive plants for each site as

$$\frac{\left(H_{\mathrm{a}}-H_{\mathrm{n}}\right)}{\left(H_{\mathrm{a}}+H_{\mathrm{n}}\right)},$$

where H _a is the Shannon diversity index of invasive flowering plant species and H _n is the diversity of native and endemic flowering plant species. The plant invasion level ranged from −1 to 1, where −1 indicated an absence of invasive species (H _a = 0) and 1 an absence of native flowering plant species (H _n = 0). The index was then averaged between the two sites for each invasional context.

To record the diversity of pollinators (Appendix S1b), we set camera traps (Ricoh WG‐70/80) at each site, positioning them in front of one flower cluster for each plant species in bloom. We selected at least one individual from every flowering plant species. If cameras remained available after covering all species, we repeated the process with additional individuals, prioritizing those with the highest flower abundance. Cameras were set to a 2‐min interval‐shooting mode for 24 h, resulting in an average of nine cameras and ∼200 sampling hours (excluding periods lost to logistical problems) per site per sampling period in May and July (May number of cameras x̄ = 9.16 [SD 2.30], recording time x̄ = 216.00 h [73.51]; July: x̄ = 8.75 cameras [2.83], x̄ = 210.08 h [71.51]). In September, sampling effort was reduced because there were fewer flowering species (x̄ = 4.5 cameras [1.05], x̄ = 103.76 h [36.50]). However, sampling effort per plant species remained consistent because September had approximately half the number of flowering species of May and July. On average, five plant species were blooming per site per sampling period (May: x̄ = 5.53 [2.10]; July: x̄ = 4.92 [2.00]; September: x̄ = 3.00 [1.30]). In total, we collected 6630 h of camera‐trap censuses (Appendix S3). Pollinators were identified from images by visual screening (Appendix S4) and classified by expert taxonomists based on the Ogasawara entomological reference collection at Tokyo Metropolitan University (TMU). When species‐level identification was not possible, individuals were classified into morphotypes (Appendix S5). We calculated diversity (H _a) of flowering plants and pollinators per site and averaged these values across sites for each invasional context and sampling period.

To estimate the abundance of anoles, we conducted 1‐h visual surveys at each site from 10:00 and 11:00 once for each sampling period. The observer spent approximately 20 min walking each 100 m of the site, visually searching for anoles from ground level to the canopy (2 h per invasional context and season). Anole abundance per invasional context was averaged over the two sites for each sampling period. Honey bee abundance was recorded using 10‐min focal censuses on two individual plants exhibiting the highest floral abundance within each site. The number of honey bees observed on each plant was summed, resulting in a total sampling effort of 20 min per site per survey. We then calculated abundance per invasional context by averaging over the two sites for each sampling period. Honey bee abundance obtained from focal censuses was significantly correlated with estimates derived from camera trap recordings (r = 0.52, df = 37, p < 0.001).

Plant–pollinator interactions

We recorded plant–pollinator interactions for all flowering plant species in each site via direct observations. During each sampling period, three observers using binoculars (Pentax Papilio) simultaneously recorded visitation events on different plants in the same site from 08:30 to 15:30. Each flowering plant was observed in a random order for 15 min with the aim of maximizing the number of individual plants observed for each species. Sampling effort was similar for all plant species, approximately 2 h per plant species at each site and sampling period. We had 420 h of plant–pollinator censuses overall. Observers focused on a flower cluster and recorded all legitimate flower‐visiting animals. We considered legitimate visitation events those in which visitors touched reproductive parts of flowers, irrespective of their potential pollination effectiveness. Therefore, we use the term pollinator to refer to all flower visitors that made legitimate flower visits. Because collection of specimens in Ogasawara is highly restricted due to the protection status of the islands, unidentified pollinators were photographed for identification by expert taxonomists. When species‐level identification was not possible, individuals were classified into morphotypes (Appendix S5). We pooled observed pollination interactions of the two sites per invasional context and sampling period.

Plant–pollinator network metrics

We built a quantitative plant–pollinator interaction network for each invasional context of each island at each sampling period with number of visits as interaction weight. To evaluate network completeness, we calculated the expected number of pollinator species with Chao's richness estimator and generated accumulation curves in relation to the number of observed interactions for each network (Chacoff et al., 2012). For each network, we calculated three indices at the network level: Shannon diversity of interactions (H), network specialization (H ₂ʹ), and web asymmetry (AS) (Blüthgen et al., 2006; Dormann et al., 2009).

At the species level, we calculated two specialization metrics for flowering plants and pollinator species in each invasion context and sampling period: number of effective partners (eH) and the degree of complementary specialization (dʹ). The number of effective plant–pollinator partners quantifies the number of partners for a given species, assuming that all plant–pollinator interactions are equally common (Dormann, 2011). This measure reflects the variety of partners a species interacts with in a given community and therefore the niche breadth of the species (Blüthgen et al., 2006). The degree of specialization describes how strongly a species deviates from a random sample of interaction partners, assuming that all species interact according to their total interaction frequencies. This metric reflects the differentiation of resources or partners among coexisting species and therefore provides information on the niche partitioning of a species (Blüthgen et al., 2006). The degree of specialization ranges from 0, an opportunistic species overlapping in its resource use with many other species, to 1, a perfect specialist species with an exclusive set of partners (Dormann et al., 2009). These species‐level metrics are robust to changes in network size, asymmetry, and number of interactions and hence are particularly useful for comparing communities with differing species richness (Blüthgen et al., 2006). We calculated network metrics with the network‐level and species‐level functions in the R package bipartite (Dormann et al., 2009).

Statistical analyses

To visualize differences in each of our nine response variables across invasional contexts, we used generalized linear mixed‐effects models (GLMMs). The variables included measures of biological invasion (plant invasion level and abundance of anoles and honey bees), community composition (plant and pollinator diversity), and pollination network structure (effective number of partners and degree of specialization for both plants and pollinators). First, we conducted five GLMM analyses for the three variables related to invasive species (plant invasion level, anole abundance, and honey bee abundance) and the two community composition variables (flowering plant diversity and pollinator diversity).

In all five GLMMs, we included invasional context and sampling period as fixed factors and site as a random factor. Sampling period was treated as a fixed factor to control for temporal variation because it had only three levels and was not appropriate as a random effect. Plant invasion level was rescaled to a 0–1 range with the linear transformation (1 + X) / 2. The transformed variable was then analyzed using a GLMM with a beta distribution, which is appropriate for modeling continuous, bounded response variables.

Next, we ran four additional GLMMs for pollination network metrics, including the number of effective partners and degree of specialization for both flowering plants and pollinators. In these models, invasional context and sampling period were treated as fixed effects, and network size was a random effect to account for differences in diversity between islands. For all GLMMs, model assumptions were checked using model diagnostics tests (dispersion and outliers) and a Kolmogorov–Smirnov test for observed versus simulated residuals and by examining QQ plots of the residuals. Tukey post hoc tests were computed to check for differences between invasional contexts within and across islands. We used Mann–Whitney tests to conduct an independent comparison between the two sites on Anijima (ANI_I and ANI_F) and to isolate the potential effect of anole presence in all the response variables. Results from this comparison were qualitatively consistent with those obtained from the GLMMs (Appendix S6).

To differentiate direct and indirect effects of different invasive species on plant–pollinator interaction networks, we used structural equation models (SEMs) implemented in the R package lavaan (Rosseel, 2012). An SEM is well suited for studying complex ecological systems because it provides a framework to simultaneously analyze direct and indirect relationships among variables (Grace et al., 2012). We structured the models around the hypothesis that invasive species affect pollination directly, by altering resource use, and indirectly, by altering plant and pollinator community diversity.

For these analyses, we first defined an a priori structure of the SEM (Appendix S7). This included all biologically plausible links between the invasive species, the plant and pollinator communities, and the final ecosystem function (effective partners and degree of specialization). We fitted a total of four separate SEMs, one for each network metric. Prior to analyses, we transformed all variables to z values (mean = 0 [SD 1]) to enable comparison of effect sizes across predictors measured on different scales. Multicollinearity among predictors was discarded because the resulting variation inflation factor (VIF) values were <3, indicating low collinearity (Zuur et al., 2010).

Based on the full a priori SEM, we performed a stepwise removal of nonsignificant paths to simplify the SEM (Vollstädt et al., 2020). Once we had our preselected variables for the initial SEM, we ran the analysis with the Satorra–Bentler correction for nonnormal distributions, excluding paths with the highest p value (Satorra & Bentler, 1994). Satorra–Bentler's (1994) approach does not allow bidirectional paths; thus, reciprocal effects are omitted. We repeated this procedure until we reached the most parsimonious SEM for each response variable, which we identified using the information theoretic approach (i.e., lowest AIC value) (Shipley, 2013). The SEMs were evaluated using standard fit criteria: χ² test, root mean square error of approximation (RMSEA), and comparative fit index (CFI). Good model fit was indicated by a nonsignificant χ² test (p > 0.05) (Byrne, 2016), a lower 90% confidence interval of RMSEA approaching 0 (Grace et al., 2012), and a CFI >0.95 (Hu & Bentler, 1999). All analyses were carried out in R 4.2.1 (R Core Team, 2022). Finally, to investigate the independent effects of invasive species on diversity and network metrics, we selected the best model based on six additional GLMMs (details in Appendices S8, S9, & S19).

To identify vulnerable taxonomic groups of pollinators, we compared species richness and number of interactions between heavily and lightly invaded islands. For this purpose, we calculated the log response ratio (LRR) as the natural logarithm of the ratio of species richness and number of interactions between the heavily invaded and lightly invaded islands: LRR = ln [(Anijima + Ototojima) / (Chichijima + Hahajima)] (Osuri et al., 2020). This analysis compared all invasional contexts of heavily invaded islands (Chichijima and Hahajima, n = 8) with those of lightly invaded islands (Anijima and Ototojima, n = 6) for each sampling period (see Appendix S10 for island‐level contrast). To avoid artifacts in the LRR due to log transformation of zeros, we calculated the minimum absolute ratio (Osuri et al., 2020) of observed values between the heavily and lightly invaded islands (Appendix S19).

RESULTS

Overview of plant–pollinator interactions

We recorded 4174 plant–pollinator interactions involving 115 pollinator species and 55 flowering plant species across the four islands (Figure 2; see Appendix S1 for list of species). Ants were excluded from the main analyses because, despite evidence of effective pollination in some temperate and arid systems (Ashman & King, 2005; Das & Das, 2023; Gómez et al., 1996), their pollination effectiveness is generally low, particularly in tropical regions (Dutton & Frederickson, 2012), though analyses including ants yielded consistent results (Appendix S11). Species accumulation curves showed that all networks were well sampled and that the number of pollinator species followed a saturating trend for all islands (Appendix S12). Plant–pollinator interaction networks showed lower complexity in heavily invaded islands—as indicated by a lower diversity of interactions, network specialization, and web asymmetry, especially in Chichijima, where anoles and honey bee abundance were highest (Figure 2).

FIGURE 2.

Quantitative plant–pollinator interaction networks across island and invasional contexts in the Ogasawara Islands (CHI, Chichijima; HA, Hahajima; CHI_N and HA_N, natural forest; CHI_D and HA_D, disturbed forest; ANI, Anijima; ANI_I, anole‐invaded; ANI_F, anole‐free; OTO, Ototojima; left column, plant species; right columns, pollinator species; line thickness, proportional to interaction frequency; red, invasive species [honey bee in case of pollinators]; BP, Bidens pilosa; SJ, Stachytarpheta jamaicensis; SM, Schima mertesiana; CS, Clinostigma savoryanum; PX, pollinator Papilio xuthus); H ₂, Shannon diversity index of interactions; H ₂ʹ, network specialization; AS, web asymmetry).

Variation in invasive abundance, community diversity, and interactions across Islands

From an invasional context perspective, the strongest contrasts occurred across islands, rather than within islands (Figure 3; full results in Appendices S13 & S14). The GLMMs indicated that plant invasion levels and the abundances of invasive anoles and honey bees were significantly higher on the heavily invaded islands (particularly Chichijima) (Figure 3a–c). Flowering plant and pollinator diversity was relatively consistent among and within islands (Figure 3d,g), although pollinator diversity was higher (yet not significantly) in the anole‐free sites than in anole‐invaded sites of the lightly invaded islands. Regarding pollination network metrics, plants had significantly more effective partners and a higher degree of specialization in lightly invaded islands (Figure 3e,f). From the pollinator's perspective, effective partners and degree of specialization remained stable across and within islands (with the exception of a significant difference in pollinator specialization between HA_N and ANI_F) (Figure 3h,i).

FIGURE 3.

(a) Plant invasion level, (b) anole abundance, and (c) honey bee abundance across invasional contexts and islands in Ogasawara and number of effective partners and degree of specialization (dʹ) for (d–f) plants and (g–i) pollinators (CHI, Chichijima; HA, Hahajima; CHI_N and HA_N, natural forest; CHI_D and HA_D, disturbed forest; ANI, Anijima; ANI_I, anole‐invaded; ANI_F, anole‐free; OTO, Ototojima; whiskers, SE; generalized linear mixed‐effects models: *p < 0.05; **p < 0.01; ***p < 0.001). Island invasional contexts (y‐axes) are presented in decreasing order of invasion category from left to right and with decreasing color saturation. Abundance measures are log transformed. Appendices S13 and S14 contain statistical results.

Effects of invasions on plant and pollinator communities and their interactions

Invasive plants and animals had a stronger direct effect on pollinator diversity (R ² = 0.40–0.50) than on flowering plant diversity (R ² = 0.09). The diversity of flowering plants was negatively associated with plant invasion level (β = −0.20, p < 0.01), but positively related to honey bee abundance (β = 0.30, p < 0.001). Pollinator diversity had the strongest negative relationship with anole abundance (plant model β = −0.59, p < 0.001; pollinator model β = −0.71, p < 0.001), followed by plant invasion level (both β = −0.20, p < 0.001) and the diversity of flowering plants (β = −0.38, p < 0.001). In contrast, honey bee abundance was positively related to pollinator diversity (plant β = 0.28, p < 0.01; pollinator β = −0.43, p < 0.01). Plant invasion level, which was weakly explained by the data (R ² = 0.10 ‐ 0.17), was positively associated with honey bee abundance (plant β = 0.41; pollinator β = 0.32, p < 0.001). However, this was largely driven by bee interactions because an additional test showed that native bees interacted significantly more with native plant species than with invasive plants (Mann–Whitney test U = 123, p = 0.005), whereas invasive honey bees did not show a clear preference (U = 57.5, p = 0.29) (Appendix S15).

Similar results were obtained for SEMs of plant effective partners (R ² = 0.15) and plant specialization (R ² = 0.30), which in both cases showed a good fit to the data (nonsignificant result of the χ² test; lower 90% of confidence intervals of RMSEA close to 0; CFI > 0.95) (Figure 4a) and were directly and positively associated with the diversity of pollinators (β = 0.22, p < 0.001). In addition, all three biological invasion factors had a negative direct or indirect effect on plant interaction metrics. Honey bee abundance was directly and negatively related to plant effective partners and specialization (β_eH = −0.25, p < 0.001; β _{d ʹ} = −0.47, p < 0.001), whereas anole abundance (β = −0.13) and plant invasion level (β = −0.04) were indirectly and negatively related to both variables through the reduction in pollinator diversity.

FIGURE 4.

Summary of structural equation models illustrating positive (black) and negative (red) relationships between biological invasion factors, community composition, and effective partners and specialization (dʹ) of (a) plants and (b) pollinators (numbers next to arrows, standardized path coefficients; *p < 0.01; **p < 0.001; ***p < 0.0001; dashed arrows, indirect effects; R ² values, best‐fit models). Structural equation models (SEMs) for effective partners (indicated as eH) and specialization (dʹ) are combined in one figure because most paths have the same coefficient. When no precision is given, the same coefficient applies to eH and dʹ. Nonsignificant paths are omitted.

The number of effective partners for pollinators was not associated directly or indirectly with any of the biological invasion or community composition variables (Figure 4b). However, the abundance of anoles was marginally associated with pollinator specialization (β = 0.12, p < 0.01), though this model had limited explanatory power (R ² = 0.01) (Figure 4b). Our best‐fit SEMs for pollinators showed only a weak fit to the data (both cases: χ² > 0.05; lower 90% of confidence intervals of RMSEA close to 0.1; CFI > 0.85) (Figure 4b) due to the lack of relations with the final response variable.

Vulnerability of functional groups

The diversity of pollinators was not associated with island size (Figure 3g), and several functional groups showed notably higher richness on lightly invaded islands, irrespective of their geographic area (Figure 5a,b). The LRR further identified vulnerable functional groups, particularly native bees, beetles (Coleoptera), and crickets (Orthoptera), which exhibited markedly reduced richness on heavily invaded islands compared with lightly invaded islands (native bees: x̄ = 1.00 [SD 0.00] vs. 8.33 [1.52]; Coleoptera: x̄ = 1.33 [0.57] vs. 6.00 [1.73]; Orthoptera: x̄ = 0.00 [0.00] vs. 1.00 [0.00]) (Figure 5a,b). Eight of the nine endemic bee species were absent from heavily invaded islands, contributing to a pronounced decline in interaction frequencies among these groups.

FIGURE 5.

Estimated average mean and standard deviation from log response ratio per functional group of (a) pollinator species richness and (b) number of plant–pollinator interactions between the heavily invaded (Chichijima and Hahajima) and lightly invaded islands (Anijima and Ototojima) (positive bars, higher species richness or interactions in lightly invaded islands; whiskers, SE). The order Hymenoptera is depicted separately for honey bees (red), native bees (dark blue), and wasps.

DISCUSSION

The invasive predatory anole had a more pronounced effect on pollinator diversity than competition from invasive bee or plant species, highlighting the importance of the invader's trophic level in determining relative ecological impact. The disappearance of key pollinator functional groups on heavily invaded islands indicates that the pollinator community was more sensitive to invasion than the plant community. The decline in native bees in concert with the presence of highly abundant honey bees affected plant visitation, reducing both the plants’ number of effective partners and specialization in these islands. Together, these findings suggest that invasion impacts in the Ogasawara Islands are predominantly driven by top‐down effects, with plant reproduction potentially threatened by predator‐driven declines in pollinator populations.

Effects of biological invasions on flowering plant diversity

Both plant invasion level and honey bee abundance had contrasting impacts on flowering plant diversity. First, plant invasion level had a direct and negative influence on the diversity of flowering plants, suggesting displacement of native flora in highly disturbed areas (Figures 3 & 4). Although flowering plant diversity did not differ across islands, plant invasion level was dependent on human activity. The two inhabited islands (frequently linked to the mainland by shipping) accounted for almost half of the overall invasive flowering plant diversity. Previous studies have shown that animal‐pollinated invasive plants can exert pressure on native plant communities, displacing them through competition for pollinators and subsequently diminishing their reproductive success (Cox, 1983; Magrach et al., 2018; Traveset & Richardson, 2014). Invasive plant species accounted for a substantial proportion of total pollinator visits, ranging from 26% to 51% (Appendix S16). One example of how invasive plants might shift the resource use of native pollinators is the butterfly Papilio xuthus, which preferentially visits the invasive species Bidens pilosa and Stachytarpheta jamaicensis over native plants, including the endemic Schima mertesiana and Clinostigma savoryanum (Figure 2).

Second, honey bee abundance was positively associated with both flowering plant diversity and plant invasion level. Previous studies in Ogasawara reported a strong association between honey bees and the invasive plant community (Kato et al., 1999) due to their generalist foraging behavior (Hung et al., 2018; Weaver et al., 2022). Although we did not observe a significant overall preference of honey bees for invasive plant species, there were indications of bee preference. For example, on the heavily invaded island of Chichijima, B. pilosa accounted for the 23% of total interactions yet represented only 9% of flower abundance (Figures 2; Appendix S17).

Preferential interactions among invasive species can lead to “invader complexes” (D'Antonio & Dudley, 1993), which facilitate the establishment of both invasive partners, reciprocally increasing their invasiveness and amplifying their ecological impacts on native communities—a process known as “invasional meltdown” (Simberloff & Von Holle, 1999). In New Zealand, for example, reciprocal facilitation has enhanced the establishment of invasive pollinators (bumble bees and honey bees) and European weeds (D'Antonio & Dudley, 1993; Goulson, 2003). Our results illustrate how interactions between honey bees and the invasive plant community, combined with the decline of native pollinators, might facilitate plant invasions and accelerate the replacement of endemic flora in the Ogasawara Islands.

Effects of biological invasions on pollinator diversity

Anoles exhibited the most pronounced negative effect on pollinator diversity; anole‐free areas (northern Anijima and Ototojima) exhibited the highest pollinator diversity. These results suggest that anoles are the main driver of the decline in the pollinator community, potentially due to their generalist arthropod diet (Losos, 2011), which includes bees, flies, and butterflies in Ogasawara (Makihara et al., 2004; Morooka, 2023), all of which have important functional roles as pollinators. The introduction of invasive predators to islands has a more pronounced impact than invasive competitors on the decline and extinction of native arthropod species (reviewed in Sax & Gaines [2008]). Our results support this pattern and provide the first quantitative evidence that anoles are the principal factor responsible for the decline in insect pollinators in the Ogasawara Islands.

The presence of invasive competitors also affected pollinator diversity, which was negatively influenced by plant invasion level. Native pollinators preferred native flowering plants, regardless of the availability of flowers from invasive plant species. This explains the negative relationship between plant invasion level and pollinator diversity, and is in line with previous studies in Ogasawara (Abe et al., 2011). The preference for native plant species by native pollinators can likely be attributed to the long‐term coevolution of the plant–pollinator network, driving selection for small and inconspicuous flowers in native plant species (Abe, 2006), different to those observed in invasive plants in Ogasawara. This raises the possibility that native pollinators could be displaced in areas dominated by invasive plant species due to the absence of their preferred floral resources. Similarly, native plant species may suffer from less effective pollination service due to the loss of native pollinators (Appendix S17). Contrary to our expectations, honey bee abundance was positively associated with pollinator diversity. Invasive species often alter community structure by shifting patterns of numerical dominance and reducing evenness (Hillebrand et al., 2008; Woitke & Dietz, 2002). Consistent with this view, studies from regions with high honey bee densities have generally reported declines in native pollinator abundances (Garibaldi et al., 2021; Paini, 2004; Wang et al., 2024). In the Ogasawara Islands, however, the observed positive relationship may reflect the relatively small scale of apiculture, which likely prevents introduced honey bees from becoming numerically dominant. Under these conditions, honey bees appear to contribute to pollinator richness without substantially disrupting community evenness, thereby increasing overall diversity, particularly on heavily invaded islands (Appendix S18). Although key variables affecting pollinator decline were included, additional links or factors not accounted for in this study such as pesticides or herbicides may also influence the pollination system.

Effects of biological invasions and community composition on interaction networks

All three invasive groups (anoles, honey bees, and plants) negatively affected plant effective partners and degree of specialization, either directly or indirectly. The direct negative effects of honey bees may be due to their generalist and intensive foraging behavior (Hung et al., 2018, 2019). This foraging behavior can lead to diet shifts in native pollinator assemblages due to competition for resources, which can negatively affect the reproductive success of native plants (Aslan et al., 2019; Magrach et al., 2017). However, competitive interactions between native and invasive species rarely cause extinctions (Sax & Gaines, 2008); rather, predators and their associated pathogens are considered a more likely invasive threat (Davis, 2003). In fact, predation by anoles exerted a substantial indirect influence on the number of effective pollinator partners and specialization of the plant community, illustrating how the top‐down effects of predation can cascade through trophic chains. Specifically, our LRR results showed how the heavily invaded islands of Chichijima and Hahajima experienced a strong decline in native pollinators. Indeed, these islands have completely lost the smallest endemic bees (<13 mm; Figueira et al., 2023), and only the largest native species Xylocopa ogasawarensis remains in situ (body length = 22–27 mm; Sakagami, 1961; Appendix S1b). This drastic reduction of native pollinators, associated with the presence of anoles, creates the conditions for honey bees to act as the dominant visitors to flowers in heavily invaded islands. On Ototojima the endemic bees Lithurgus ogasawarensis and Ceratina boninensis mainly pollinate Hibiscus glaber, whereas on Chichijima, where these bees are locally extinct, flower visits are dominated by honey bees. Consequently, native plant species that have lost their native pollinators in these islands may benefit to some extent from the presence of honey bees. In the aftermath of native species extinctions, the introduction of invasive species with the same functional role can act as a mechanism of biotic compensation (Gurevitch & Padilla, 2004; Thomsen et al., 2017). Nevertheless, the effectiveness of honey bees as pollinators of native plants is not guaranteed (Magrach et al., 2018; Travis & Kohn, 2023).

Although the decline in pollinator species richness is associated with reduced floral visitation to the plant community on heavily invaded islands, plant reproduction remains a complex process because visitation does not necessarily equate to effective pollination. Pollination effectiveness is strongly influenced by visitation rates but also depends on factors such as conspecific pollen deposition and time spent on the flower (Schupp et al., 2017; Tierney et al., 2023). We used visitation rates as a proxy for pollinator efficiency because we could not directly measure plant fitness or pollinator performance across the large plant–pollinator community, which somewhat limits inference about effects on plant reproduction. However, a previous study on heavily invaded islands in Ogasawara (Abe et al., 2011) showed that honey bee visitation rates were correlated with fruit set, though many native and invasive plants still experienced significant pollen limitation (50% of native and 53.3% of endemic species). This implies that honey bee visits cannot completely compensate the loss of native pollinators and that there may be a reduction in the success of sexual reproduction of native plants on heavily invaded islands of Ogasawara. Further studies are required to quantify honey bee pollination efficiency, especially with respect to fruit and seed set and viability.

Neither biological invasions nor community composition were associated with the number of effective partners and specialization for pollinators. This pattern suggests high functional redundancy, where pollinator generalization, common in island pollination systems and among successful invaders (Traveset & Richardson, 2014), may buffer networks against changes in plant or pollinator diversity and composition, allowing pollinators to maintain a stable number of interactions despite local species losses (Bascompte & Scheffer, 2023). These results highlight that plants and pollinators may respond differently to invasion impacts, emphasizing the importance of considering both network components when assessing ecological consequences of invasions.

In summary, our results revealed that the combined effects of an invasive predator and competitors pose a significant threat to the native fauna and flora of the Ogasawara Islands, particularly native bees and the ecological functions they provide. In a simultaneous process, the anole may disrupt the animal‐mediated pollination of native plant species by preying on native pollinators, whereas the invasive honey bee promotes pollination of certain invasive plants not visited by native bees (Abe et al., 2011). Species heterogeneity is crucial for ecosystem resilience (Bascompte & Scheffer, 2023), and our results revealed a depleted pollinator community with the disappearance of endemic bees and depletion of beetles on heavily invaded islands. The number of partners and the degree of specialization are fundamental characteristics of interaction networks and are intricately tied to the stability of ecological communities (Estes et al., 2011; Gonzales & Arcese, 2005; Schleuning et al., 2015). Thus, the strong reduction in these factors, together with the loss of key pollinator functional groups, is indicative of ecological communities facing severe biodiversity loss. This reduction in diversity threatens ecosystem resilience by diminishing the redundancy of ecological functions provided by native pollinators (Albrecht et al., 2013; Menke et al., 2012), potentially leading to broader ecological disruptions.

Our findings have broad implications for invasive species management and the conservation of native species, demonstrating that biological invasions must be evaluated jointly to capture the complex dynamics governing ecological interactions networks. We found that the consequences of species introductions are context dependent and contingent on the trophic level of the invader, the nature of their interactions (predation or competition), and, importantly, the number of coinvasions because impacts can have synergic effects that further unbalance (invasional meltdown) or rewire (native plant pollination compensation) pollination interaction networks. This multi‐invasion framework provides a pioneering quantitative assessment of simultaneous coinvasions, revealing the dominance of top‐down effects and their cascading influence through trophic chains. These insights underscore the particular urgency of strengthening biosecurity and early‐detection programs on islands where ecological networks are exceptionally vulnerable, extinction risks are disproportionately high, and invasive species remain the leading driver of biodiversity loss. Reliance on eradication alone is insufficient, often ethically challenging, and typically implemented too late to safeguard the ecological functions most at risk.

AUTHOR CONTRIBUTIONS

M.Q., I.P.‐S., and A.L.C. conceived the original idea. M.Q. designed research with input from I.P.‐S., A.L.C., S.M.T., and A.T. M.Q., M.M., I.P.‐S., S.M.T., and A.L.C. conducted field work. M.Q. curated and analyzed data. M.Q. wrote the paper with input from I.P.‐S. and A.L.C. All authors contributed to the final manuscript.

Supporting information

Supporting Information

Quitián, M. , Planas‐Sitjà, I. , Morooka, M. , Traveset, A. , Tierney, S. M. , & Cronin, A. L. (2026). Coinvasional disruptions to island pollinator networks. Conservation Biology, 40, e70304. 10.1111/cobi.70304

DATA AVAILABILITY STATEMENT

The original dataset is available at https://doi.org/10.5281/zenodo.19221853.