Colonization and coexistence of non‐native ants on a model Atlantic island

Colonization by non‐native ants represents one of the gravest potential threats to island ecosystems. It is necessary to identify general mechanisms by which non‐native species are able to colonize and persist in order to inform future prevention and management. We studied a model‐island assemblage of 17 non‐native ant species with aim of identifying the spatial source of introductions and assessing how such a diversity of species are able to coexist.


| INTRODUC TI ON
Non-native ants are among the most damaging of invasive taxa.They have caused billions of dollars of damage annually to agriculture, public and social welfare sectors across the globe (Angulo et al., 2022) and frequently impact health (Gruber et al., 2022).Beyond human costs, invasive ants can instigate widespread ecological collapse and even cause species extinctions (O'Dowd et al., 2003;Wetterer & Espadaler, 2021).Oceanic islands have historically been especially vulnerable to ant invasions (McGlynn, 1999;Suarez et al., 2009).
Such islands support disproportionately high levels of biodiversity and species endemism (Kier et al., 2009;Veron et al., 2019), and island-endemics are also disproportionately impacted by invasive species (Clavero et al., 2009).Non-native ants are therefore of particular threat to the unique and irreplaceable biodiversity of islands.
The documented specific impacts of invasive ants on island biodiversity are numerous.The infamous yellow crazy ant, Anoplolepis gracilipes, has been shown to disrupt frugivory by Christmas Islandendemic birds (Davis et al., 2010) and separately out-compete native hermit crabs for carrion resources on the Tokelau archipelago (McNatty et al., 2009).On the Hawaiian Islands, introduced ants have been observed reducing fledging success of native seabirds (Plentovich et al., 2009) and are suspected of driving the extinction of the common Pacific skink, Emoia impar (Fisher & Ineich, 2012).
The invasive big-headed ant, Pheidole megacephala, has been shown to predate endemic achatinellid snails on the Ogasawara Islands of Japan (Uchida et al., 2016).Impacts extend to native island ants; P. megacephala has likely caused the extinction of the Cape Verdeendemic ant Monomorim boltoni (Wetterer & Espadaler, 2021).Ant invasions are therefore proven to reduce island biodiversity and ecosystem processes through a range of direct and indirect means.
Effectively conserving island biodiversity will require both management of past ant colonization and prevention of new introductions.Ants are well-adapted to be transported by humans (Suarez et al., 2009;Ward et al., 2005) and continental cities have been shown to act as entry points for non-native species (Brassard et al., 2021).
Understanding if or how introduced species spread from smaller settlements will be key to reducing their spread on islands, and increased knowledge of ant natural history and behaviour will increase success in targeted eradication attempts (Hoffmann et al., 2016).
Importantly, predicted climate change is likely to globally alter habitat suitability for invasive species (Bertelsmeier et al., 2015).This means that timely filling of knowledge gaps presents significant potential for limiting climate-driven shifts in ant invasion patterns.Some especially useful studies on ant invasion have focused on the process of dispersal through native habitats.Pietrek et al. (2021) quantified the rate at which P. megacephala invaded savanna habitat in East Africa, concluding a value of approximately 50 m per year.In a similar study of A. gracilipes, Abbott (2006) found supercolonies on Christmas Island to be expanding at up to 50 cm per day through the native rain forest.These findings represent opportunity for predicting when specific ant species will reach important spatial milestones in contiguous habitat, and planning local interventions accordingly.
However, because of differences in species ecology, such estimations are of limited value to broader-scale applications concerning multiple ant species and diverse habitats.
Other important studies have identified the mechanisms by which introduced ants exploit available niche space in order to establish and coexist.Ecological theory dictates that no two species can coexist in the same niche space (Hutchinson, 1959), but that space can be partitioned through various mechanisms.Lester et al. (2009) studied the coexistence of 17 non-native ant species on Tokelau and found negative species co-occurrence relationships at island scale.They concluded that ant species were achieving coexistence by limiting inter-specific competition via habitat separation.Coexistence is also possible via fine-scale resource partitioning.Balzani et al. (2021) tested this hypothesis by conducting stable isotope analysis to detect shifts in trophic niche of the ant Formica paralugubris outside of its native range.The authors found that the species exhibited considerable plasticity in its diet, apparently in order to avoid niche overlap with native ants.A third coexistence hypothesis is that environmental variation prevents competitive exclusion in diverse assemblages (Andersen, 2008;Hutchinson, 1961).Gippet et al. (2022) found that native and invasive ants were able to coexist around buildings as a result of daily microclimatic variation.
In complex ecosystems, it is likely that diverse ant communities are maintained by a combination of each of the three mentioned coexistence mechanisms, although we can find no evidence that this has previously been tested.
Identifying predictable patterns in ant colonization and niche partitioning is vital for assessing the vulnerability of island ecosystems and designing management strategies.Studying such ecology has previously been difficult.Studies on diverse non-native assemblages in particular are lacking, as are studies investigating the historical processes of ant invasion.Valid and useful research would depend on extensive standardized sampling replication at islandscale, simple yet heterogeneous ecosystems from which generalizations could be extrapolated, and accessible species records dating to early in the ant introduction process.
We argue that Ascension Island in the South Atlantic Ocean represents an ideal model system for studying non-native ant invasion.
The volcanic island is just one million years in age (Jicha et al., 2013) and isolated from the closest continent (West Africa) by 1600 km.As a result of its biogeography, the island has no obviously native ants which could confound study of the non-native species (Appendix S1; Wetterer et al., 2007), has low native diversity in other taxonomic groups (Ashmole & Ashmole, 1997;Duffey, 1964) and has no meaningful biological exchanges with other islands beyond humanmediated species introductions.The island is 88 km 2 -an area of relevance to other inhabited oceanic islands around the world but that is small enough to survey in its entirety.It is compact in shape and therefore minimizes edge effects of coastline.Various ecological habitats span the island area, ranging from barren lava flows in the dry lowland to cloud forest in the mountainous centre (maximum elevation: 859 m).Importantly, there also exist accounts of non-native ant occurrence dating back to the late 1800s (Dahl, 1892).
On this model island, we aimed to identify mechanisms of ant colonization which would serve as general rules-of-thumb applicable to conservation of island biodiversity globally.We questioned: do non-native ant species spread at a general linear rate, and do they radiate from small human settlements on islands?We then asked: how do multiple ant non-native species coexist over the moderate to long term (>50 years)?We aimed to assess whether each of (1) islandscale habitat separation between species, (2) separation in fine-scale foraging behaviour, and (3) climatic heterogeneity were maintaining high diversity in the Ascension ant assemblage.Our hypothesis was that a combination of all three mechanisms was responsible.

| ME THODS
We addressed our aims by collecting ants across Ascension Island in a highly standardized procedure.Sampling occurred in two parts: firstly through collection of ants around human settlements every 3 months between 2019 and 2022, and secondly through intensive survey of the whole island between January and April 2022.After collection and taxonomic identification of all ants, we consulted the scientific literature to establish first known record of each species and assess rates of spread.
The first part of our sampling involved visiting sites within human settlement every March, June, September and December (Figure 1).These 10 sites were selected for being close to storage areas for inbound cargo and were therefore intended to represent potential origin sites of ant introduction.On each site visit, eight open 10 mL sample tubes were positioned in ant-accessible areas.A range of baits were deployed in order to attract diverse ant species of various feeding ecology, and also to represent a range of high-sugar, high-fat and high-protein resources broadly equivalent to common foods that could be reasonably implied to be available to newly introduced ants during transport and establishment.The first of these was processed meat in water; an animal protein source with high fat content.The second was unsweetened peanut butter; a non-animal protein source with similarly high fat content.The third and fourth were strawberry jam and marmalade, representing low-fat, highsugar resources of different fruit origin.Two baited tubes of each were deployed at every site.After 1-2 h, the tubes were closed and removed for analysis.We did not count the ants inside each tube, we only recorded species presence.
For the second part of our sampling, we divided the island into Universal Transverse Mercator grid cells of 1 × 1 km (UTM zone 28 South) and sampled as many of those cells as possible within our available study window (Figure 1).Cells were selected and visited randomly.Within each cell, we sampled three sites best-representing the diversity of habitats in that 1 km 2 area.
In order to detect ants of various ecology and behaviour, we employed several survey methods at each of those sites (Figure 2).At all sites, we deployed four pitfall traps in a square formation with sides of approximately 25 m.Pitfalls comprised a transparent 200 mL capacity plastic cup dug flush into the substrate.The cup was filled with 50 mL of water mixed with detergent in order to break the surface tension.
Once filled, ground pitfalls were covered with a white-painted square of plywood of size 20 × 20 cm in order to shelter the trap from rain and reduce evaporation of the liquid.Where there were trees present, a further four "hanging pitfalls" were deployed on trees closest to each vertex of the square.Those hanging pitfalls were larger at one pint capacity (568 mL) and filled with 100 mL of the same detergent solution.They were hung at approximately 1.5 m elevation in order to sample from the crown of the mostly low-growing trees on Ascension.Pitfalls were deployed for 3 days before collection.
Each ground pitfall, and hanging pitfall where present, was baited according to one of four treatments.Pitfalls and hanging pitfalls were paired such that each vertex of the square formation received the same treatment.Baits varied from prior surveys in settlements in order to better represent the different scavenging resources expected to be available across Ascension's landscapes.One in four of the vertices were not baited, to act as a control comparison.Two further were baited with strawberry jam and processed meat, as per earlier trials.These were chosen for their high-sugar and high-fat contents over marmalade and peanut butter respectively, as the latter two baits were proved less effective at attracting diverse ants in the earlier surveys.The fourth vertex was baited with spotted moray eel flesh; a native and lean protein source representing fish carcasses frequently discarded by seabirds.All baits were suspended above the trapping liquid inside a filled 3 mL receptacle derived from a disposable pipette bulb.We emphasize that datasets derived from using both the earlier and later set of baits were used to determine species point occurrences; however only the later-collected control/ jam/meat/fish-bait data was used in addressing our hypotheses on the possible mechanisms of niche partitioning.
Within the square formation, we actively searched for ants using a further range of methods targeting all present vegetation and detritus.At all sites, we searched for 15 min using an aspirator to collect ants.Where low, leafy vegetation was present, we performed F I G U R E 1 Ascension Island, showing elevational gradient, buildings and settlements (light blue), settlement sampling locations (black dots) and 1 km 2 UTM grid cells.Grid cells which were randomly selected for surveying (separately from within-settlement surveys) are bordered in red, while cells which were not selected are grey.and contacted the small number of invertebrate specialists who had visited the island within that period to assess whether further species had since been detected.For each known species, date of earliest collection was recorded.Species for which we recorded the first Ascension record were attributed to the relative survey year (2019-2022).These earliest records were utilized as best-possible estimations of actual colonization.
All statistical analysis was conducted in R version 4.2.2 (R Core Team, 2022), and all data were plotted using the package ggplot2 (Wickham, 2016).Approximate island-scale Extent Of Occurrence (EOO) for each species was estimated via convex hulls encompassing all sites at which the species was detected across all sampling methods.Species which were found only within settlements were given an area value of 0. Convex hulls are known to over-simplify occupancy and be poor estimators of species rarity (Irl et al., 2017); however, we rely on them at this stage to purposefully simplify the process of island-scale range expansion and derive basic rules of thumb on local EOO increase on islands.It is important to note that species with large EOO may still have low area of occupancy, and thus be "rare" (Irl et al., 2017).Site-level habitat heterogeneity within EOO boundaries is therefore crucial in explaining species occupancy.This is addressed in the later analysis centred on habitat partitioning.
We used linear regression to quantify the strength of the relationship between dates of first species record and island EOO.For each sampled site outside of settlements, we calculated the distance to the closest building using a complete infrastructure map supplied by the Ascension Island Government (unpublished).With this information, we conducted a mixed effect logistic regression using the lme4 package (Bates et al., 2015).We modelled the presence/ absence of ants outside of settlements (coded as 1's and 0's respectively) as a function of distance from building and year of first species record, with differences between ant species accounted for via inclusion of a per-species random intercept.
We conducted all further analyses on niche separation via our three proposed mechanisms using only the longest-established ant species, that is those for which we found an earliest record of 50 years prior or more.In order to determine spatial habitat separation between ant species, we performed Pearson's correlation tests on per-site presence/absence species data.Positive correlation indicated preferential coexistence of species at specific sites, while negative correlation indicated habitat separation.
Correlations were calculated only within overlapping convex hulls of species to remove sites where one or neither of the compared species had been able to expand into yet.In addressing our hypotheses on niche separation, further study of specific perspecies habitat usage was relevant only if habitat separation was detected.
We continued to quantify the importance of the other two mechanisms, fine-scale resource usage and climatic variability, using only data from the pitfalls (both ground level and hanging).We attempted to quantify a five-dimensional foraging niche space for each of the longest-established ants by combining measures of perspecies abundance, vertical space usage, and attraction to each of the fish, jam and meat baits.To do so, it was necessary to derive site-wise indices of each on axes of comparable scale.Abundance, A, was calculated simply as the base-10 logarithm of the total number of each species trapped in pitfalls at each site.Vertical space usage, V, was calculated as: where N Hi is the number of occupied hanging pitfalls at site i and N Gi is the number of occupied ground pitfalls (both out of four).V i is bound between −1 and 1, such that −1 represents exclusive trapping in ground pitfalls and 1 represents exclusive trapping in hanging traps.
Bait attraction, B, was calculated as: where n xi represents the total number of individuals caught in traps baited with x, n 0i is the number of individuals trapped in non-baited pitfalls, and n i is the total number of individuals caught in pitfalls at site i.
Similar to the previous index, −1 represents complete avoidance of the bait while 1 represents complete and exclusive attraction to the bait.
Five-dimensional hyper-volumes representing fine-scale foraging niche spaces were approximated for each species using the dynRB package (Schreyer et al., 2022).Each of A, V, B F (fish attraction), B J (jam attraction) and B M (meat attraction) were combined in dynamic range boxes with 101 steps.Overlaps between per-species range boxes representing niche spaces were measured as proportions based on the default "product" method of variable aggregation.We tested whether niche overlaps were greater or lesser than expected by chance by fitting 999 randomized sets of range boxes, in which the species names in the input dataset had been shuffled, and combining with the one true fitting to generate distributions of median overlap from which p-values could be derived.
Finally, we assessed the importance of daily climatic heterogeneity using redundancy analysis (RDA).We requested and processed daily weather data collected by the UK Met Office on Ascension (unpublished).The data included daily measures of wind speed (knots), humidity (%), temperature (°C), rain showers and cloud cover (oktas), from which we derived mean wind speed, humidity, temperature and cloud cover over three-day trapping periods as well as total number of rain showers.We also computed distance-based Moran's eigenvectors (dbMEMs) using the adespatial package (Dray et al., 2022) in order to describe the positive spatial autocorrelation between ant communities at different sites.We Hellinger-transformed the total per-site abundance data using the vegan package (Oksanen et al., 2022) and selected via step-wise permutation tests both the weather variables and dbMEMs which explained a significant amount of variance in the resulting community matrix.Using the selected dbMEMs, we partialled out the effect of positive spatial autocorrelation before computing a final RDA constrained on only the selected weather variables.

| RE SULTS
We collected and identified 14 species from 47,473 ants and 63 sites allocated around Ascension, as well as three further species from the additional 10 sampling locations inside human settlements (Table 1).The 17 total species collected constituted all nine named species previously confirmed to be established on Ascension (Wetterer et al., 2007).Three of the remaining eight could be assigned to genus level only and were Monomorium sp. 1, Solenopsis sp. 1 and Solenopsis sp. that one of the unidentified Solenopsis species that we detected is the same as "Solenopsis sp.1" -the tenth species listed by  1, of which we collected single individuals only.

Species
The greatest calculated convex hull area belonged to P. megacephala, which was the earliest recorded species (Figure 3; Table 1; Dahl, 1892;Wetterer et al., 2007).We found a very strong negative linear relationship between year of first record and hull area (Figure 4a; p < .001).The linear regression suggested that each ant species increased in EOO by 0.51 km 2 per year (95% CI: 0.38-0.64km 2 per year) since introduction.The very high amount of variance explained by the model (R 2 = 0.80) suggested that the relationship held relatively constant between different ant species.
We also found significant linear relationships between species detection and distance from the closest building (Figure 4b), but only where there was an interaction term with year of first record included (Z = 2.78, p < .01).In this model, the distance term was also significant (Z = −2.48,p < .05)but year of first record alone was not (Z = 1.42, p > .05).The significant interaction indicated that recently introduced species were detected more frequently nearby to buildings, but species which were long-established were more frequently detected away from buildings.
Our data highlighted only limited correlation in occurrence between the ant species first recorded over 50 years ago: C. emeryi, M. subopacum, P. alluaudi, P. longicornis, P. megacephala and S. globularia.There were no significant negative correlations in occurrence (Figure 4c).The majority of correlations were positive, although only the correlations in occurrences of C. emeryi with P. alluaudi (ρ = .52,p < .05)and P. megacephala (ρ = .48,p < .01)were significant.This suggested that non-native ants were not coexisting via habitat separation on Ascension, and in fact some species preferentially occupied sites which were already occupied by others.and S. globularia (p < .001)were trapped more frequently at ground level compared with tree level (Figure 5b).Only P. alluaudi (p > .05) showed no significant preference between vertical strata.
Bait attraction differed between species (Figure 5c-e The data suggested that M. subopacum was the relative generalist of the six species, foraging at intermediate abundance, also with no bait preference. The differences between species in foraging behaviour manifested in low niche overlap (Figure 6).Species-by-species quantification of overlap showed interactions in which niche overlap was significantly lower than expected by chance (Figure 6a).Redundancy analysis of ant community composition highlighted differences in species behavioural responses to weather, facilitating further niche separation (Figure 7).Stepwise selection of terms showed each of cloud cover, humidity, number of rain showers and air temperature to each explain a significant amount of the variance in composition, along with nine dbMEMs explaining positive spatial autocorrelation (p < .05).Those 13 variables together explained 41% of total variance.After the variance attributable to the dbMEMs had been conditioned out, the first two RDA axes represented 90% of explained variance.
Most of the long-established species appeared to be heavily influenced by daily weather conditions.Hellinger-transformed P. megacephala abundance loadings were most strongly correlated with humidity loadings (ρ = .47).In direct contrast, S. globularia loadings were negatively correlated with humidity loadings (ρ = −.83).
P. alluaudi loadings were correlated strongly with cloud cover (ρ = −.42),indicating that the species is most active on bright days.
Only P. longicornis showed weak relationships with weather patterns; the strongest correlation between loadings was with cloud cover (ρ = .16).Of the five, it could therefore be inferred that P. megacephala and S. globularia exploit opposite ends of the humidity scale, while C. emeryi, M. subopacum and P. alluaudi activity are separated more so by temperature and sun exposure.ant introductions by exploiting their ecology (Hoffmann et al., 2016).
We found that non-native ants were generally colonizing island area at around 0.5 km 2 per year.Although there was some expected variation between species in comparison of the areas that they had colonized (20% of variance was unexplained), that overall model was strong.Besides the species' differing ecology, perhaps some variation could be attributed to the long periods between early surveys.
For example, assuming unrealistic perfect detection, species first detected in 1958 could have been present for over 65 years before being recorded (the prior survey was 1892).In any case, our estimated rate was comparable to those derived from individual species by other authors (Abbott, 2006;Pietrek et al., 2021), but only at small spatial scales.Other measures have mostly been in terms of linear distance, which require conversion in order to derive island-scale estimates.Using simple trigonometry, a circular colonized area of 1 km 2 would be expanded annually by 19% according to Pietrek et al. (2021, studying P. megacephala) and 75% by Abbott (2006, studying A. gracilipes).Our equivalent rate of 51% sits firmly between those estimates.Conversely, in consideration of a circle of 44 km 2 (half of the area of Ascension), we predict an expansion rate of just 1% compared with 3% (Pietrek et al., 2021) and 10% (Abbott, 2006).This is clearly in part due to differences in rate quantification and inherent biases in how earlier estimations upscale.However, there is potential that our low expansion rate at relatively large spatial scales could be explained partially by density-dependent processes.Further investigation of long-term and long-distance range expansions in non-native ants is certainly warranted to address this hypothesis.
In agreement with past studies on large mainland cities (Brassard et al., 2021), we confirmed that ant colonization on our model island was stemming from buildings and small settlements, with newly introduced species being confined nearby-to or within those areas.
Three species were found only within settlements (E.latinodes, M. pharaonis and N. bourbonica).Of those, at least M. pharaonis is very closely linked to homes and buildings (Wetterer, 2010).This suggests that this species is unlikely to expand in EOO across the island and may eventually disappear.In view of this, M. pharaonis and other species closely linked to settlements may be hypothesized to be less of an ecological threat to native habitats than other ants.Unfortunately, though, M. pharaonis has potential to cause significant damage through other means, for example as a vector of human illnesses (Wetterer, 2010).For this reason, non-native ants confined to settlements should still be considered high-risk.The identification of settlements as entry points to island ecosystems suggests key management actions which could be taken in order to reduce ant introductions.First and foremost, biosecurity measures should be stepped-up at all airports and boat ports on vulnerable oceanic islands, with aim of preventing the humanfacilitated movement of ant species into settlements in the first place.In particular, training in ant detection and identification should be prioritized for all island-based biosecurity workers, who should be provided with the taxonomic resources and microscope access required for the task.Secondly, we recommend natural habitat restoration surrounding island settlements which may include propagation of native plants.Habitat quality is often poorer in close proximity to settlements (Ahrends et al., 2010;Xiong et al., 2020) and on Ascension this manifested in dominance of invasive vegetation, mostly comprising Mexican thorn (Prosopis juliflora; Ascension Island Government, 2015).Invasive ants are known to establish more readily on barren and disturbed habitats (Bernal & Espadaler, 2013;Ramalingam & Dharma, 2022), and so restoring such habitat may create an ecological barrier to ant establishment outside of settlements.Such habitat restoration effort to meaningful buffer widths would be costly, but undoubtedly cheaper than the comparative cost of managing continual invasive ant colonization in the future.In addition, habitat restoration will increase the population of native biodiversity and improve the ecological processes in long term.
Our results showed that non-native ant species achieve optimal niche separation through both resource partitioning and climatic variation.C. emeryi found distinct niche space by targeting our lean protein baits at low abundance, thus avoiding competition with the abundant and aggressive P. megacephala, and doing so during periods of low sun exposure and air temperatures.M. subopacum also avoided high temperatures, but was more generalist in foraging behaviour than C. emeryi.P. alluaudi foraged in lowest abundance and made more equal usage of ground-and tree-level strata than other species, while uniquely being most active on clear days.P. longicornis avoided competition with M. subopacum, P. alluaudi and P. megacephala through the species' strong attraction to sweet baits at ground level.P. megacephala exhibited the greatest overlap on foraging niche space of other species and appeared to compete by foraging at great abundance in humid conditions.In contrast to this behaviour, S. globularia exploited arid weather, choosing to forage predominantly at ground level and target lean protein sources.
These findings supported part of our initial hypothesis; however, we had also expected to detect niche separation through differential habitat usage also.We found no evidence of habitat separation, in contrast to Lester et al. (2009).This might be explained by the lack of ecosystem complexity on Ascension in relation to other older islands, although our study incorporated significant gradients in elevation, microclimate and habitat structure.This explanation therefore seems unlikely, especially in comparison with the Nukunonu islands of Tokelau (Lester et al., 2009) which are just 5.5 km 2 in total area and <5 m in maximum elevation.We instead suggest that our statistical control for ongoing spatial colonization (using only overlapping convex hull areas) may have removed the artificial signal of habitat separation.Studies elsewhere may be confounded without such data treatment.
Our results did support the conclusions drawn by Balzani et al. (2021) and Gippet et al. (2022) on resource-partitioning and climate heterogeneity, respectively.While we found considerable evidence that both processes promote non-native ant coexistence on islands, we also found that not all studied species were impacted by each.P. longicornis was shown to significantly avoid resource competition, but did not appear to be affected by weather.Conversely, M. subopacum was most active during periods of cool weather, but exhibited no preference for food resources.Following this reasoning, non-native ant species elsewhere may partition habitat usage also, however we must conclude that resource-partitioning and climate variability are the dominant processes which maintain diversity.
These species-level differences in niche partitioning at least partially confirm our hypothesis that multiple mechanisms drive coexistence in non-native ant assemblages.
The dividing of niche space at fine scale, rather than via habitat separation, raises major warnings of future ant colonization.Our analyses strongly suggest that non-native ants are able to coexist at high diversity within very small areas.Because of this, it is likely that non-native ants can potentially persist at diversities far greater than we currently observe on oceanic islands, and may be able to saturate available niche space through domination of resources and microhabitats.For this reason, we stress that prevention of future colonization is critical to avoid negative impacts on native biodiversity, even on small islands where many non-native ant species are already established.
Having said that, the invasive ecology that we describe here could also be exploited through informed management.We found that non-native food resources divided behaviour of ants and promoted coexistence.Control methods on islands should focus on removing as many of those resources as possible.Non-native protein sources could be reduced by eradication of introduced vertebrate species, in turn removing foreign organic matter in the form of corpses and animal waste.Some non-native ants are closely associated with non-native scale insects which exchange their honeydew secretions in exchange for ant protection (Davis et al., 2010).Control of, for example, introduced scale insects would therefore reduce the sugar-based resources available to ants.Likewise, eradication of fruiting non-native plates would reduce foreign sugar resources.
Ant dependence on climatic heterogeneity may represent an alternative method of control.We found that the majority of nonnative species in our study were strongly associated with foraging at ground level.Stabilizing microclimate at this stratum may then be key to reducing available niche space.Microclimate beneath native vegetation is more stable compared with heavily disturbed habitats (Hardwick et al., 2015).As a result, restoration of native vegetation presents significant scope for reducing ant niche partitioning via climate.
There exists considerable potential for targeting multiple nonant species with local eradication by identifying key sites at which they coexist.This could be through chemical control, which is currently being trialled on the Atlantic island of Saint Helena.On for an additional 15 min.Where there existed significant leaf litter or loose top soil, we collected three 1 × 1 m samples into a canvas bag and extracted ants in the lab using custom-made Berlese funnels positioned underneath a 35 w halogen bulb for 3 days.At all sites with trees, we collected three vegetation-beating samples by placing a 1 × 1 m white cotton sheet underneath three sturdy branches and firmly hitting the branch five times with a metal pole to dislodge foraging ants.After assembling a list of ant species present on the island, past records of each were identified from the scientific literature.We relied heavily onWetterer et al. (2007), who themselves collated and confirmed all known ant records from Ascension before 2007.We searched extensively for published records spanning 2007-2022 Sampling design at each site visited outside of settlements, with specific pitfall methods shown.Each 25 × 25 m sampling area was defined by four standard pitfalls placed at the vertices (red spots).Where trees were present, further hanging pitfalls (green spots) were positioned as close as possible to each vertex.Hand collection and, where habitat structure permitted, sweep-netting, litter-sampling and vegetation-beating were performed within each sampling area.
2. As described in Wetterer et al. (2007), the nine previously identified species were first recorded by either Dahl (1892), Duffey from his 1958 survey (1964), Ashmole & Ashmole from their 1990 survey (1997), Cutler & Szal from their 2002 survey (unpublished) or Mendel from his 2003 survey (unpublished).It is likely, though unconfirmed, Wetterer et al. (2007), and the only which could not be assigned a true identification.We found no evidence of ant species which were newly recorded between 2003 and 2022.The most numerous species in our surveys was P. megacephala, with 30,797 individuals.The least were Nylanderia bourbonica and Monomorium sp.

F
Ascension Island (grey) showing sampling sites outside of settlements (crosses).Convex hulls representing EOO of three example species are shown: Pheidole megacephala, recorded first in 1892, Monomorium subopacum, recorded in 1958, and Solenopsis sp. 2, recorded in 2022.Spatial arrangement of non-native ants on Ascension Island.(a) Significant positive linear relationship between per-species year of first record and local EOO, based on convex hulls.Grey band represents 95% confidence interval.(b) Significant linear relationship between site distance from nearest building, year of first record and perspecies presence/absence coded as binary values.Both axes are square-root transformed.Points were given jitter in the y-direction for the purpose of clarity.Coloured bands represent 95% confidence intervals.(c) Correlations in site presence/absence between all of the longest-established species (recorded 1958 and earlier).Blue represents positive correlation, and "+" represents positive correlation where p < .05.Red represents negative correlation, and there were no negative correlations where p < .05. b o p a c u m P .a l l u a u d i P .l o n g i c o r n i s P .m e g a c e p h a l a From the ground pitfall and hanging pitfall data, we detected niche separation between the six longest-established species (Figure 5; n = 41,223).P. megacephala foraged in greatest abundance with a median of 226 individuals per occupied trap (Figure 5a).P. alluaudi foraged in lowest abundance with a median of just 1. Wilcoxon tests indicated that each of C. emeryi (p < .001),M. subopacum (p < .001),P. longicornis (p < .001),P. megacephala (p < .001) C. emeryiexhibited reduced overlap the niche of P. megacephala (p < .05),with lowest proportional overlap on the dimension describing foraging abundance (overlap = 0.16).P. longicornis similarly exhibited reduced niche overlap onto the niche space of P. megacephala (p < .05)with low overlap on the dimension describing attraction to fish bait (overlap = 0.18).P. longicornis also showed reduced overlap with M. subopacum (p < .05)and P. alluaudi (p < .05),both with lowest per-dimension overlap in fish attraction (overlap = 0.15 & 0.13 respectively).Finally, S. globularia exhibited reduced overlap on the niche space of P. alluaudi (p < .05)with minimum overlap also in fish attraction (overlap = 0.14).Median niche space overlap between all species was significantly lower than expected by chance when quantified via all of the three aggregation methods calculated by the dynRB package(Schreyer et al., 2022; Figure 6b; p < .01).

F I G U R E 5
Foraging behaviours of each of Ascension Island's longest-established non-native ants.Colours represent different species.(a) Abundance: the total number of individuals trapped per-site.(b) Vertical usage: −1 represents complete and exclusive ground usage and 1 represents complete and exclusive foraging usage of trees per-site.(c-e) Fish/ jam/meat attraction: −1 represents total avoidance of the bait while 1 represents total attraction.
Our results from Ascension Island highlight key concepts of ant invasion ecology which have global significance to the conservation of island biodiversity.This model system, comprising 17 species introduced over more than 130 years, demonstrates new and unexpected mechanisms of coexistence between non-native ants.From the presented data, we warn of the potential for continued ant colonization long into the future, but suggest cost-effective methods for managing

F I G U R E 6
Niche overlap in foraging behaviour of Ascension Island's ants.(a) Species-by-species visualization of the proportion of niche space for each Overlapped Species shared by the Overlapping Species.Overlap values significantly lesser than in 1000 randomized datasets (p < .05)are represented by "−."There were no cases where overlap was significantly greater than in those same randomized datasets.(b) Averaged niche overlaps across all species via three methods: Product, Mean and Geometric Mean(Schreyer et al., 2022).In each, the true median is represented by a vertical red line, while the densities of values generated from 1000 randomized datasets are shown in grey.Weather effects on ant community composition on Ascension Island.RDA scaling is type 1. Cloud cover, humidity, rain shower frequency and maximum air temperature each explained a significant amount of variance in the Hellinger-transformed community data.Coloured arrows represent loadings for each species.