Trait filtering during exotic plant invasion of tropical rainforest remnants along a disturbance gradient

1. Human-modified tropical landscapes are often invaded by exotic plant species, but relatively few species are able to colonise remnant areas of rainforest embedded within such landscapes. The functional traits of successful invaders of natural versus anthropogenic habitats are poorly known, especially in tropical regions, and identifying such traits provides insight into the mechanisms that drive invasion. Here, we examine the invasion of tropical rainforest remnants along a disturbance gradient, within a human-modified agricultural landscape, and determine whether exotic species that invade these forests are selected according to particular traits. 2. We surveyed the occurrence of 18 exotic species along 100-m transects in four habitats—oil palm road, forest-oil well as traits that are more to of native forest species tall and woody), making these exotic species more able to compete and persist in that environment. Our results show that disturbed tropical rainforests with open canopies are susceptible to invasion and highlight the traits of exotic species which can invade rainforest habitats, and which may pose a threat to regenerating tropical rainforests.


| INTRODUC TI ON
Understanding the processes driving plant invasion within different ecosystems has been a long-standing goal of ecological science (Elton, 1958;Lodge, 1993). As human population and land-use change increases, it is vital to understand these processes in humanmodified landscapes. Successful invasion of any habitat requires species to overcome sequential ecological barriers separating different stages of invasion: introduction, establishment, spread and impact (Levine, Adler, & Yelenik, 2004). Life-history and functional traits of invading species are influential at each stage of invasion and play an important role in key invasion hypotheses (Catford, Jansson, & Nilsson, 2009). As such, identifying the traits of successful invaders provides insight into which mechanisms are driving invasion and can facilitate predictions of whether an invading species will be successful in a particular habitat.
There is no set of traits universal to all invading species and across all invaded habitats, as ecological barriers faced by exotic species vary between ecosystems (Pyšek & Richardson, 2008). However, exotic species will readily spread into new areas if they have traits associated with efficient dispersal (e.g. many seeds produced regularly, small seed mass, long-distance dispersal; Colautti, Grigorovich, & MacIsaac, 2006;Nathan et al., 2008;Pyšek & Richardson, 2006;Rejmánek & Richardson, 1996). Once introduced, invading species will only establish persisting populations if they are adapted to the environmental conditions of the invaded habitat and can compete successfully with the native species community, and those that are not adapted will be filtered out (Procheş, Wilson, Richardson, & Rejmánek, 2008;Weiher & Keddy, 1995). The traits which facilitate invasion in one habitat may not confer the same outcome in another due to differences in environmental conditions (Sher & Hyatt, 1999). In disturbed habitats, the increase in resource availability promotes invasion (Blumenthal, 2006;Davis, Grime, & Thompson, 2000) as the majority of invasive species are light-loving ruderal species (i.e. r-strategists; Pyšek & Richardson, 2008;Rejmánek & Richardson, 1996). Therefore, many exotic species are unlikely to invade undisturbed environments where resource availability (i.e. light and space) is low, including closed-canopy forest ecosystems. In both temperate and tropical forests, only those exotic species that have some degree of shade tolerance and traits that allow them to compete with the native flora will establish, which could for example include high relative growth rate, large above-ground biomass, high maximum height and high photosynthetic rates (Bufford, Lurie, & Daehler, 2016;Koike, 2001;Pyšek & Richardson, 2008;van Kleunen, Weber, & Fischer, 2010;Weiher et al., 1999). Thus, exotic species that are invasive in highly disturbed anthropogenic habitats may not necessarily be able to colonise adjacent less-disturbed, closed-canopy natural habitats.
The spread of exotic species into natural tropical forest habitats from invaded anthropogenic areas is not frequently considered in the invasion ecology literature as tropical forests have been considered to be robust to plant invasions thus far (Fine, 2002).
However, deforestation across the tropics, due to recent expansion of large-scale agricultural practices, has resulted in exotic plant species being frequently introduced by humans via newly created transport networks (Hulme, 2009;Levine & D'Antonio, 2003;Seebens et al., 2015). These heavily invaded agricultural areas may act as sources of exotic propagules that disperse into areas of native forest retained within these landscapes as conservation set-asides . However, traits that result in successful invasion of open agricultural areas are likely to differ from those that result in successful invasion of natural forest habitats. In this study we investigate, for the first time, the spread of exotic plants along a gradient of disturbance from highly invaded, anthropogenic habitats (oil palm plantations) into native tropical rainforest remnants that are recovering from disturbance, and test whether there is selection for particular traits across the disturbance gradient.
The abiotic conditions and biotic communities of native rainforest remnants are very different to agricultural areas (Laurance et al., 2002), and so the anthropogenic habitat edge is likely to act as a strong ecological barrier to invasion. However, within tropical forest remnants themselves, the forest structure and community can be highly heterogeneous due to a combination of edge effects  , 2011). These disturbances result in physical gaps in the canopy, which increase light levels and thus facilitate the establishment of introduced exotic propagules (Döbert, Webber, Sugau, Dickinson, & Didham, 2017a;Peters, 2001;Teo, Tan, Corlett, Wong, & Lum, 2003;. Many exotic species are not well adapted to closed canopy, intact forests, however, and habitat filtering may mean that only a few shade-tolerant species present in the landscape can establish and persist in these environments (Dawson, Burslem, & Hulme, 2009;Koike, 2001;Martin & Marks, 2006).
In many parts of South-East Asia, land-use change has resulted in a highly fragmented landscape with disturbed rainforest remnants surrounded by predominately oil palm plantations, particularly in Malaysia and Indonesia (Gaveau et al., 2014). These landscapes provide a useful system to study exotic species that have established populations in the oil palm matrix and examine how they disperse and colonise within remnant forests (i.e. conservation set-asides). The understorey of oil palm plantations are known to be dominated by exotic plant species (Fee, Tui, Bin, & Hoy, 2017;Rembold, Mangopo, Tjitrosoedirdjo, & Kreft, 2017), therefore the threat of invasion into areas of remaining forest is high due to this constant source of propagules. It is important to determine the vulnerability of rainforest remnants to invasion because they have been shown to be crucial for maintaining local and regional biodiversity (Hill et al., 2011), ecosystem services (Koh, 2008;Ricketts, 2004) and carbon storage  within sustainable agricultural tropical landscapes. Therefore, disentangling the mechanisms that drive invasion in disturbed forests will help in the management of human-modified tropical landscapes, as well as advance our knowledge of plant invasion processes within tropical forests.
In order to investigate the processes driving plant invasion of tropical rainforest remnants, we compare species and trait compositions of exotic species found along a gradient of disturbance within a human-modified agricultural landscape in Sabah, Malaysian Borneo. Disturbance and light levels decrease along this gradient from oil palm agricultural areas to closed canopy, intact forest, and so we expect exotic species richness and occurrence to decrease accordingly. We then specifically examine functional traits that address the barriers exotic plants encountered during the colonisation of new environments (i.e. invasion), including dispersal (dispersal syndrome and seed mass) and persistence (potential competitive ability: maximum plant height, longevity: stem density and seed production: seed mass ;Weiher et al., 1999). Based on previous studies, we expect the exotic community inside the forest to have traits that indicate efficient dispersal and allow propagules to readily spread from oil palm source populations into rainforest remnants (i.e. long-distance dispersal and small, light seeds), as well as traits that promote the establishment of persisting populations (higher maximum height, woody habit) and populations may either benefit from efficient dispersal of smaller seeds or resource provision of larger seeds.

| Study sites and transect surveys
Exotic plant communities were studied over two field seasons (July-October 2017 and February-April 2019) in six large-scale oil palm plantations in Sabah, Malaysian Borneo, which had remaining areas of native forest embedded within them (Figure 1). These forest conservation set-asides (hereafter termed 'rainforest remnants') had usually been left unplanted because they were unsuitable for oil palm cultivation (e.g. too steep or rocky) and had been commercially logged (with varying intensity, e.g. due to differences in topography), prior to plantation development, probably between ~1991 and 2009 (Reynolds, Payne, Sinun, Mosigil, & Walsh, 2011). Most of these rainforest remnants are currently managed retrospectively as sites of 'high conservation value' (HCV; Senior, Brown, Villalpando, & Hill, 2015), and therefore should not have experienced significant ongoing direct human disturbance as indicated by the certification process for sustainability developed in 2005 by the Roundtable on Sustainable Palm Oil (RSPO, 2018). Across the six plantations, exotic species were surveyed at 21 sites, in areas of oil palm abutting rainforest remnants.
The majority of these rainforest remnants (n = 15 sites) were fully surrounded by oil palm, with six sites connected to a larger area of forest reserve outside the plantation boundary. Even though these sites were connected to other forest patches, these six sites were predominately surrounded by oil palm and had experienced the same disturbance histories as the other sites (see Table S1 for individual site details). Sites were at least 1 km apart to avoid spatial autocorrelation.
At each study site, 100-m transects were surveyed in up to four different habitats (Figure 1c), to capture variation in exotic species occurrences dispersing from oil palm planted areas to forestoil palm edges and then into disturbed logged forest (hereafter 'disturbed forest') and intact logged forest (hereafter 'intact forest'). The four habitats differed in terms of light levels and past and current disturbance (Table S2 and Figure S1). Oil palm transects (n = 21) were located along the roadsides of major unpaved plantation roads. Forest-oil palm edge transects (n = 21) were either along small unpaved roads (motorbike traffic only) or along oil palm terraces abutting forest. Disturbed forest transects (n = 20) were located on average 53 m from the closest forest-oil palm edge, along old logging skid trails, abandoned oil palm terraces or unpaved roads (no longer in use). They were in highly degraded forest habitat that varied in terms of canopy openness and understorey ground cover. Intact forest transects (n = 16) were on average 90 m from the closest edge and were considered to be structurally intact (i.e. closed canopy forest), which was usually on steep slopes with a generally sparse ground cover. Due to previous disturbance (i.e. logging and clearing) and most remnants being small (<100 ha; Table S1) the majority of these forests are likely to have extensive edge effects that extend throughout the rainforest remnant (Lucey et al., 2017). Access inside the forest was not possible at one of the 21 sites because it was too steep, and there was no forest site that was deemed to be 'intact' in five very small (<12.5 ha) and highly disturbed rainforest remnants. In oil palm planted areas, active weed management, including herbicide spraying and vegetation cutting, occurred predominantly around the base of oil palm trees. We do not expect these localised management practices to lead to major changes in species composition within the surveyed parts of the oil palm habitat, because our oil palm transects were carried out away from the bases of oil palm trees, in areas which were not sprayed or cleared recently.
Along each 100-m transect, the presence/absence of exotic plant species was recorded within every 1-m section and 1 m either side of the transect (i.e. 2 m 2 sections; Figure 1d). The presence of exotic species in each 1-m section of the transect was our measure of occurrence (i.e. range = 0-100 per species per transect). Pilot transects during the first field season were used to identify exotic species known to be present in Sabah, based on species list produced from literature and online resources (see Bakar, 2004;CABI, 2019;Döbert et al., 2017a;Fee et al., 2017;Pallewatta, Reaser, & Gutierrez, 2003;Peh, 2010), and any additional suspected exotic species encountered in the field (over both field seasons), were identified and their exotic status in Sabah confirmed with the help of local botanists (Danum Valley Research Centre and Sandakan Herbarium). We did not include a leguminous cover crop (Calopogonium mucunoides) that was intentionally planted in the oil palm areas, exotic graminoids due to identification challenges, nor any morphospecies that could not be identified to species.
This left 18 exotic species, which we believe to be representative of the exotic plant community found within these landscapes.
In order to determine variation in the canopy structure and light levels between and within the four habitats, canopy cover was recorded every 20 m (five points in total per transect, with the mean of four measurements facing each cardinal direction per point) using a concave spherical densiometer (www.fores try-suppl iers. com). To examine the complexity of the forest structure, the number of large native trees [>20 cm diameter at breast height (DBH)] along the transect was recorded every 10 m, 5 m either side of the transect and not including oil palm trees. Mean canopy cover (%) and total number of large native trees were used as measures of forest openness and canopy complexity at the transect level.

| Functional traits of exotic community
Trait data for the 18 study species were obtained from online databases (predominately species datasheets in the Invasive Species Compendium (CABI, 2019) and Kew Seed Database (Royal Botanic Gardens Kew, 2019) but see Table S1 for additional sources) for traits that address potential barriers to colonisation (Table 1).
Functional traits were selected for inclusion in the analysis if there were data on at least 75% of the study species (Table 1).
This resulted in four species-level traits for analysis: plant maximum height (indication of potential competitive ability), plant woodiness (proxy for stem density), seed dry mass and dispersal syndrome (including both primary and secondary dispersal modes; F I G U R E 1 Study location on Borneo (a), within the Malaysian state of Sabah (b), showing the 21 study site locations (black circles), and areas of primary and logged forest and industrial oil palm plantations (Gaveau, Salim, & Arjasakusuma, 2017). (c) Drone image of site F2, showing transect locations (i.e. four habitats): 1 = oil palm, 2 = forest-oil palm edge, 3 = disturbed forest, and 4 = intact forest. Oil palm = major plantation roadside; Forest-oil palm edge = minor plantation road or terrace where oil palm meets forest; Disturbed forest = skid trail, old road or abandoned terrace; Intact forest = closed canopy forest. (d) Transect sampling. Transects were 100 m × 2 m with presence/absence recorded for each exotic species (n = 18) in each 1-m (2 m 2 ) section along the transect see Table S3 for species-level trait expressions). Average functional trait values per species were calculated for seed dry mass when multiple data points were available. Dispersal syndromes were classified into a binary code for adaptation to long-distance (i.e. wind, water and vertebrate) or short-distance (e.g. gravity/ unspecified, vegetative propagation, ballistic or invertebrate) dispersal. Additionally, binary traits for presence/absence of vertebrate dispersal and wind dispersal were retained for analysis, to determine which long-distance dispersal mode was the most prevalent in the exotic forest community (we excluded the consideration of water dispersal due to lack of rivers in our study sites).
Note that these binary dispersal traits are non-exclusive, e.g.
Chromolaena odorata is dispersed by both wind and vertebrates.
Community weighted mean trait values (CWM; weighted by species occurrence) were calculated per transect using the fd package (Laliberté, Legendre, Shipley, & Laliberté, 2014) in r 3.4.2 (R Core Team, 2019). For binary traits (plant woodiness, long-distance dispersal, vertebrate dispersal, wind dispersal), we calculated the proportion and number of 'occurrences' (presences in 1-m transect sections) with and without the trait. Prior to CWM calculations, plant maximum height and seed dry mass were log 10 transformed to improve normality, as well as model convergence (see statistical analyses below). Covariance and phylogenetic relationships amongst traits were tested and it was concluded that these would have little impact on CWM (see Appendix S1).

| Exotic species community composition
We tested for differences in the occurrences of each study species in the four habitats using Bayesian generalised linear mixed models (GLMMs; i.e. 18 models), using the blme r package (Dorie & Dorie, 2015).
Unlike traditional GLMMs, Bayesian GLMMs can accommodate complete separation in a logistic regression (i.e. an explanatory variable that completely separates the response variable into all zeros or ones) by allowing priors on the fixed effect terms (Bolker, 2015). This was the case for occurrence data in our study, where some species were never recorded inside rainforest remnants. In our models, we used highly uninformative Gaussian priors for all fixed effects, with means of 0 and variances of 9 and no covariance between them. For each model, the response variable combined the frequencies of presence/ absence of the species in each 1-m transect section (ranging 0-100), the fixed effects were habitat (four levels: oil palm, edge, disturbed forest and intact forest) and canopy cover (as both linear and nonlinear quadratic terms to account for nonlinear individual species responses to light). Site and transect-level random effects were included to account for pseudo-replication. Species with very low total occurrence across all transects (<50 occurrences across all 78 transects) were not analysed. For each model, we tested whether geographical distance among transects was influencing our findings using Moran's I autocorrelation tests but found no evidence for geographic structuring in residuals of any of the models (Appendix S1).
To compare compositional similarity of exotic species across the four habitats we conducted an ordination using non-metric multidimensional scaling (NMDS) on a Bray-Curtis dissimilarity matrix (Magurran, 2013). Ordinations were conducted on transect-level data, based upon the total occurrence of species per transect (i.e. total number of 1-m sections where the species was recorded present, range = 0-100/transect) using the vegan package (Dorie & Dorie, 2015;Oksanen et al., 2007). We assessed the marginal effect of habitat and structure (canopy cover and number of large trees) on community composition using a permutational multivariate analysis of variance (PERMANOVA), with 999 permutations to calculate significance.
TA B L E 1 List of plant functional traits linked to invasion success of rainforest remnants. Data were compiled predominately from each species datasheet in the Invasive species compendium by the Centre for Agriculture and Bioscience International (CABI; www.cabi.org/ isc/) and Kew Seed Database (data.kew.org/sid/), see Table S1 for additional sources and individual species trait expressions. The number of species for which data are available (out of the 18 species surveyed) are in parentheses. 'Predicted relationship' refers to the expected filtering during invasion from oil palm to forest edge and into disturbed forest and then intact forest, based on previous studies: (Dawson, Burslem, & Hulme, 2011;Pyšek & Richardson, 2008;Rejmánek & Richardson, 1996;van Kleunen et al., 2010;Weiher et al., 1999)

| Functional trait variation across habitats
To test for differences in each of the four functional traits among the four habitats, Bayesian GLMMs were fitted with fixed effects of habitat and canopy cover, and random effects of site and transect. Six separate models were run for the following response variables: CWMs of log 10 plant maximum height and seed dry mass (with Gaussian error distributions), and binomial proportion that were woody, long-distance dispersed, vertebrate dispersed and wind dispersed (with binomial errors). Pairwise comparisons between habitats were carried out with a Tukey post hoc test using the emmeans package (Lenth & Lenth, 2018). For each model the residuals were tested for geographic structuring and no evidence was found for any trait model (Appendix S1).
Finally, robustness of the results was evaluated by re-fitting the models without the dominant invasive species, Clidemia hirta, to determine whether relationships were maintained across the remaining exotic plant community.

| Exotic species community composition
Oil palm plantations and forest-oil palm edges were highly invaded with similar exotic species richness and occurrence ( Figure 2 and Table S4; plantation mean richness = 9.2 species/transect (SE ± 0.58), mean occurrence = 12.94/transect (SE ± 3.76); edge richness = 7.8 species/transect (SE ± 0.69), occurrence = 12.73/ transect (SE ± 3.14) respectively), although the species compositions on each transect differed significantly between the two habitats, as well as between sites within habitat ( Figure 3). Species richness and occurrence then decreased markedly in disturbed forests, where a subset of the exotic community present in the oil palm and forest edge habitats was found ( Figure 2 and Table S4; mean richness = 3.1/transect (SE ± 0.64); mean occurrence = 3.66/ transect (SE ± 0.88)), and only one species (Clidemia hirta) was found in intact forest, where it was found at low levels of occurrence (mean = 0.91/transect (SE ± 0.21)).
Whilst 18 exotic species were recorded overall, no single transect contained all 18 species (oil palm: 4-14 species per transect, Forest-oil palm edge: 2-13 species, disturbed forest: 1-10 species; Figure 2). The tropical shrub Clidemia hirta made up ~33% of all occurrences (Table S4). Some species had very low occurrence and appeared on only a few transects (e.g. Lantana camara and Crassocephalum crepidioides; Table S4), whereas some species were very common (i.e. Asystasia gangetica and Ageratum conyzoides; Table S4 and Figure S2). Of the six most abundant species, some decreased in occurrence from oil palm into disturbed forest (e.g. A. conyzoides, A. gangetica and Mimosa pudica; Table S4 and Figure S2) whereas some species became more common at forest edges than in plantations and decreased again in forest habitats (e.g. Clidemia hirta, Chromolaena odorata and Hyptis capitata; Table S4 and Figure S2).
As expected, canopy cover and number of large native trees both increased from oil palm to intact forest ( Figure 2 and Table S5).
Bayesian GLMMs for the 16 species with sufficient occurrence data to analyse, showed that the occurrence of five species (including two F I G U R E 2 Exotic plant species richness and occurrence, and local habitat variables in four habitats in Sabah, Malaysian Borneo. See main text and Table S2 for habitat descriptions of the most abundant; Clidemia hirta and Chromolaena odorata) was significantly associated with canopy cover, both with linear and quadratic terms (Table S6 and Figure S3). Thus, some species were more common at intermediate light levels (e.g. A. conyzoides, Clidemia hirta and Chromolaena odorata) and others require full light (e.g. Euphorbia hirta). Habitat remained significant for most species (n = 11) after controlling for the effect of canopy cover (Table S6), suggesting some additional environmental, spatial or biotic factors were also important for affecting the occurrences of these exotic species.

Differences between species composition based on Bray-
Curtis distance measures showed that habitat separated on NMDS ordination plots, with oil palm plantation and intact forest the most different, and edge transects intermediate (Figure 3). Both habitat and canopy cover significantly affected species composition and helped to drive the observed separation in community composition (Figure 3; PERMANOVA; habitat, R 2 = 0.1, p < 0.001; canopy cover, R 2 = 0.06, p < 0.001). The additional effect of distance to forest edge was tested for forest transects only, as a proxy for propagule pressure, and was found to have no significant effect on the occurrence of any species (Appendix S1). This confirms the role of local environmental conditions in driving observed species compositions.

F I G U R E 3
Non-metric multidimensional scaling (NMDS) of species composition based on Bray-Curtis index for oil palm, forest edge, disturbed forest and intact forest. Each symbol represents one transect and species names are in italics. Habitat is the same as described in Figure 1. Results of Permutational Multivariate Analysis of Variance, with 999 permutations were: number of large trees, p = 0.350; habitat, p = 0.001; canopy cover p = 0.001. For visualisation, random x-axis jitter has been added to the intact forest points, which were composed of a single species and so had no variation in NMDS scores

| Functional trait filtering
The exotic community showed clear trends in mean functional trait values along the disturbance gradient, despite there being different exotic source communities in the oil palm across sites (Figure 4; Tables S4 and S5). Plant height and woodiness increased, and seed dry mass decreased from oil palm into intact forest. Additionally, exotic species adapted for long-distance dispersal (i.e. vertebrate or F I G U R E 4 Trends in six functional traits of exotic species (n = 18) across three habitats from oil palm plantation into rainforest remnants with all species included (block outline) and dominant invasive Clidemia hirta excluded (dotted outline). p-values are from ANOVA, after controlling for canopy cover, on Bayesian generalized linear mixed models (***p < 0.001; **p < 0.01; *p < 0.05; NS = non-significant). Pairwise comparisons between habitats detailed in Table S8 F I G U R E 5 Trends in six functional traits of exotic species (n = 18) in relation to canopy cover, with all exotic species included (orange) and with dominant invasive Clidemia hirta excluded (grey). Regression lines are on the raw data of those significant relationships from the results of Bayesian generalized linear mixed models, after controlling for habitat. p-values are the significant results of ANOVA tests wind dispersed) were more common inside rainforest remnants. All traits were significantly related to habitat as well as to canopy cover (Figures 4 and 5; Table S7).
To test the robustness of the trait results, we removed the dominant exotic species, Clidemia hirta, from the analyses, but found this did not substantively change the main findings. The remaining exotic community (n = 17 species) exhibited increased height, woodiness and vertebrate dispersal from plantation to disturbed forest ( Figure 4; Tables S7). Whilst the direction of effect was the same, the effect of canopy cover became non-significant when C. hirta was removed ( Figure 5), apart from woodiness which remained significant.
As with the species composition analyses, distance to forest edge had no significant effect on the mean functional trait value of forest communities for any trait (Appendix S1). Our study also supported previous findings that closed canopy rainforest have low levels of invasion (Dawson, Burslem, & Hulme, 2015;Döbert et al., 2017a;Teo et al., 2003), and provided the first evidence of this result in conservation set-asides embedded within oil palm plantations. Despite most remnants being small in size (<100 ha; no. sites = 15) and the intact forests being on average <100 m from a heavily invaded anthropogenic matrix (Appendix S1) only one exotic species was found. However, we found that the most disturbed areas of these rainforest remnants are much more susceptible to invasion predominately due to disturbance history and a more open canopy, and not because they were on average closer to the forest edge and therefore sources of propagules. The number of exotic species and their occurrence decreased along the disturbance gradient (from oil palm to intact forest), which is likely partly attributed to the reduction in resource availability (e.g. light), as well as other factors that vary between habitat.

| D ISCUSS I ON
The most common species we recorded, Clidemia hirta, is known to be found only in open areas in its native range but can invade closed canopy forest throughout its introduced range (DeWalt, Denslow, & Ickes, 2004), and was the only species found in closed canopy forest in this study. Indeed, we noted that C. hirta was smaller in the very shaded parts of these forests, which may indicate growth restriction in shaded, closed canopy areas. However, we report for the first time that this species is significantly more abundant in a forest edge habitat, which we believe is due to C. hirta favouring intermediate light levels as supported by the relationship with canopy cover.

| Dispersal into rainforest remnants
We found that rainforest remnants are more likely to be invaded by exotic species with longer-range dispersal syndromes, presumably because they are better able to disperse seed inside rainforest remnants without the requirement of human assistance. Other studies have shown that landscape-scale habitat heterogeneity reduces the spread of short-distance dispersed invasive species but has little impact on species that disperse long-distances (Marco, Montemurro, & Cannas, 2011). This may explain our finding that long-distance dispersal (i.e. vertebrate, wind and water) promotes invasion in our highly heterogeneous anthropogenic landscape.
We found species that are vertebrate-dispersed (internally and externally, and both primary or secondary dispersal modes) are more likely to invade forests than those dispersed solely by wind or without long-distance dispersal adaptations. In these landscapes it is likely that small birds and bats are effective dispersers of species with fleshy fruit (e.g. C. hirta), having been identified as the main dispersers of exotic woody species in tropical rainforests (Rejmánek, 1996), as well as preferentially foraging in and adjacent to areas of recovering tropical rainforest canopy gaps (Schupp, Howe, Augspurger, & Levey, 1989). Secondary dispersal by terrestrial mammals is also likely to occur for those species with different primary dispersal mechanisms (e.g. the wind dispersed Chromolaena odorata). It is likely that In this study, we considered only natural dispersal agents as there should be no significant ongoing disturbances in these rainforest remnants. However, previous disturbance may have facilitated early propagule arrival into these forests on the machinery used during timber extraction; this has been observed as a mechanism in the introduction of invasive grasses in tropical dry forests (Veldman & Putz, 2010), as well as the role of domestic vehicles across a range of habitats (Ansong & Pickering, 2013). In addition, forest set-asides may experience low levels of human disturbance, and so propagules may be dispersed by people moving through the forest, by transporting seeds attached to clothing (Mount & Pickering, 2009).

| Persistence within forest environments
The shift to exotic species with greater maximum heights inside rainforest remnants indicates that the ability to compete for light is an important marker of success in the forest environment (Weiher et al., 1999). Other invasive exotic species have been found to be taller when compared with non-invasive exotic species (Gallagher, Randall, & Leishman, 2015). In addition, the increased woodiness (i.e. stem density) of the exotic forest community is indicative of increased longevity and so species that survive longer (i.e. perennial shrubs) are more likely to persist within the disturbed areas of these rainforest remnants than herbaceous annuals. The ability to survive and successfully compete with native forest species is necessary for persistence within new environments (Weiher et al., 1999) and there is likely greater niche overlap and therefore competition between young native trees and exotic shrubs, than small herbaceous exotic species.
The positive relationship between canopy cover and woodiness in our results (and same trend with maximum height) suggests that light is driving some of the observed patterns of trait filtering. In the more shaded forest, shrubby exotic species may be more likely to successfully compete for light with native species. Dominant herbaceous species found in the oil palm, and at the forest-oil palm edge, with the capacity for long-distance dispersal (e.g. Ageratum conyzoides) by contrast are unable to establish and persist in large numbers inside the forest. The inclusion of data on leaf functional traits could provide additional insight on the trait filtering mechanisms occurring during colonisation, especially the relationship with light, because they provide information on another dimension of resource acquisition and allocation. Specific leaf area (SLA) is also well-correlated with growth rates (Lambers & Poorter, 1992) and high SLA is often attributed to invasive species (Grotkopp & Rejmánek, 2007;Hamilton et al., 2005). Although foliar data were lacking for our species, this would be a useful focus of future work to help ascertain whole-plant trade-offs.

| Habitat and other influential factors
Our finding that trait patterns and differences in species occurrences are significantly explained by habitat may derive from differences in land-use history that impact environmental conditions other than light. For example, along the disturbance gradient there is likely to be a change in the soil conditions, with high compaction in the most disturbed areas (those outside forest and on skid trails or roads inside rainforest remnants), which is known to promote exotic plant establishment in logged tropical rainforests (Döbert et al., 2017a). In addition, increased soil compaction can have a negative impact on native species regeneration (Hattori et al., 2013) and thus reduce competition between exotic and native species. Soil pH and nutrients are known to influence exotic plant richness in tropical rainforests (Teo et al., 2003), however, a previous study in these rainforest remnants found characteristics of the soil (specifically pH and available P) to be less important for invasion than other drivers . Outside the rainforest remnants, soil characteristics are likely to be more influential on the exotic plant community due to the addition of fertilisers (that add N, P and K to the soil) and presence of leguminous cover crops which increase N availability (Huenneke, Hamburg, Koide, Mooney, & Vitousek, 1990).
The role of exotic species traits is likely to be one of several processes driving invasion within this landscape, and it is probable that other factors are also at play. Species-specific biotic interactions that vary among habitats may also be influential, as exotic species are known to both lose and gain biotic interactions when invading new areas [e.g. presence of available pollinators (Sargent & Ackerly, 2008), soil microbes (Reinhart & Callaway, 2006), damage from herbivory and pathogens (DeWalt et al., 2004)]. Therefore, specific interactions with abiotic and biotic factors not recorded in this study may contribute to individual species invasion success alongside the variables we analysed.
In addition, invasion is a dynamic process and therefore species persisting in oil palm plantations may still be expanding their populations into rainforest remnants (resulting in a time lag on colonisation) as has been recorded in temperate rainforest remnants (Essl, Mang, & Moser, 2012). Invasion time lags are thought to be shorter in tropical regions than in temperate regions, and shorter for exotic herbaceous species than woody species (Daehler, 2009

| Conservation implications and management suggestions
Our results suggest that exotic plant invasion might not be an imminent threat to native rainforest remnants if they have low levels of previous disturbance and high canopy cover. However, rainforest remnants within agricultural areas are often small and highly degraded due to disturbance from logging and extensive edge effects (Haddad et al., 2015;Lucey et al., 2017;Reynolds et al., 2011), as well as being isolated from larger tracts of primary forest. These disturbances are known to alter the native plant community to include a higher proportion of pioneer and non-forest specialists (Laurance et al., 2006;Pütz, Groeneveld, Alves, Metzger, & Huth, 2011) and isolation can reduce pollination and dispersal of native trees (Ghazoul & McLeish, 2001), both of which negatively impact the natural regeneration of forest species (Haddad et al., 2015) and may result in the intact parts of these forests becoming degraded over time. In these circumstances, exotic species may be able to spread through the forest into new gaps, using the disturbed areas of these forests (i.e. old skid trails and roads) as 'stepping stones' to invade deeper into the forest core.
Although we believe the 18 species we recorded in this study to be a good representation of the exotic plant community in these landscapes, we recognise that other species may be present in other parts of these landscapes, especially in the oil palm. In addition, native non-forest generalists that proliferate with disturbance and increased light availability (e.g. Callicarpa, Melastoma, Uncaria) were also observed in the rainforest remnants but were not recorded during this study. This native 'weedy' community may follow the same trait patterns observed in the exotic community, because native understorey tropical plants are known to become less functionally diverse inside primary rainforest when compared with repeatedly logged rainforest (Döbert, Webber, Sugau, Dickinson, & Didham, 2017b). Our results provide useful information about which exotic plant species and their associated traits, are invading these forests, including species that are known to suppress the regeneration of native species through competition and allelopathic effects (e.g. Clidemia hirta and Chromolaena odorata; Ambika, 1980;Ishmaini & Lesstari, 2015;Nakamura & Nemoto, 1993).
To help reduce invasion and thus minimise any potential negative impacts on native rainforest regeneration, as well as improve overall habitat quality for forest biodiversity, we recommend active forest restoration within canopy gaps in conservation set-asides, which may combine enrichment planting and weed removal. This would be especially beneficial in highly disturbed sites which lack closed canopy forest, in order to decrease light levels in the understorey and reduce colonisation by most exotic species. In addition, we suggest targeted removal of exotic species with the traits highlighted in this study from the source population in oil palm areas to stop or slow the spread into rainforest remnants, especially in new plantings where exotic populations may not yet have established.
Likewise, we caution against planting of exotic species for ornamental reasons, for example, the ornamental species, Lantana camara, which has the traits that would, according to our results, make it a successful invader of remnant forests (i.e. it is a vertebrate dispersed shrub). Furthermore, as the anthropogenic forest-oil palm edge is the avenue for invasion in these forests, as seen in other tropical rainforests , decreasing the edge to area ratio in rainforest remnants is likely to help minimise invasion, as well as reduce edge effects on native forest species richness and structure, which are known to be greater in irregular shaped fragments (Bennett & Saunders, 2010;Didham & Lawton, 1999). These suggested measures could be incorporated into certification procedures and documentation (e.g. RSPO) to help ensure that exotic species do not become a threat to the conservation status and long-term sustainability of these important remnants.

| CON CLUS IONS
Our study is the first to show strong evidence for trait filtering of exotic species as the environment changes from anthropogenic habitats into remnant areas of recovering native tropical rainforests along a disturbance gradient. Successful invasion of rainforest remnants requires species to have both adaptation for long-distance dispersal, in particular vertebrate dispersal, and traits that facilitate competition with native forest species (i.e. tall and woody), but there must also be a level of disturbance (i.e. canopy openness). Our results highlight both the type of forest susceptible to invasion (i.e. previously disturbed with an open canopy) and the type of exotic species that can invade remnant areas of native forest within increasingly anthropogenic tropical landscapes (Potapov et al., 2017).
These rainforest remnants maintain biodiversity within agricultural tropical landscapes, and our findings provide a first view of which exotic species invade the understorey of these conservation setasides. Although, it is not yet clear whether these exotics affect long-term regeneration of native forest communities, a precautionary approach is to introduce management practices that facilitate native seedling regeneration and increase canopy cover, to decrease the occurrence of most exotic species and reduce potential impact.