Evidence of biotic resistance to exotic plant invasion in degraded Bornean forests

Intact tropical forests are generally considered to be resistant to invasions by exotic species, although the shrub Clidemia hirta (Melastomataceae) is highly invasive in tropical forests outside its native range. Release from natural enemies (e.g., herbivores and pathogens) contributes to C. hirta invasion success where native melastomes are absent, and here we examine the role of enemies when C. hirta co‐occurs with native Melastomataceae species and associated herbivores and pathogens. We study 21 forest sites within agricultural landscapes in Sabah, Malaysian Borneo, recording herbivory rates in C. hirta and related native Melastoma spp. plants along two 100‐m transects per site that varied in canopy cover. Overall, we found evidence of enemy release; C. hirta had significantly lower herbivory (median occurrence of herbivory per plant = 79% of leaves per plant; median intensity of herbivory per leaf = 6% of leaf area) than native melastomes (93% and 20%, respectively). Herbivory on C. hirta increased when closer to native Melastoma plants with high herbivory damage, and in more shaded locations, and was associated with fewer reproductive organs on C. hirta. This suggests host‐sharing by specialist Melastomataceae herbivores is occurring and may explain why invasion success of C. hirta is lower on Borneo than at locations without related native species present. Thus, natural enemy populations may provide a “biological control service” to suppress invasions of exotic species (i.e., biotic resistance). However, lower herbivory pressures in more open canopy locations may make highly degraded forests within these landscapes more susceptible to invasion.

Several hypotheses have been proposed to understand invasion processes (see Catford et al., 2009 for an overview) and they are often context specific, differing between taxa and ecosystems (Catford et al., 2021;Jeschke et al., 2012). "Darwin's Naturalization Conundrum" captures two conflicting invasion hypotheses that center around the degree of phylogenetic relatedness between the invading exotic species and the invaded native community (Diez et al., 2008). If an exotic species has close relatives in the native community, then it is likely to be pre-adapted to the local environmental conditions due to shared evolutionary histories and phylogenetic conservatism of functional and life history traits (Flynn et al., 2011). According to this hypothesis, exotic species are unlikely to be filtered out in the early stages of invasion in regions where confamilars occur (i.e., habitat filtering hypothesis; Weiher & Keddy, 1995;Procheş et al., 2008), promoting their invasion success. On the other hand, the occurrence of closely related native species increases the likelihood of strong competition for resources with related natives and pressure from specialist natural enemies, such as herbivores and pathogens, that have evolved to feed on a particular plant family or genus (i.e., biotic resistance hypothesis; Coley & Aide, 1991;Hodkinson & Casson, 1991;Marquis, 1991;Gaston, 1993). These natural enemies can contain the spread of exotic species by limiting their growth and reproduction, as would occur in their native ranges (Levine et al., 2004). These opposing hypotheses highlight the importance of considering closely related species when studying exotic plant invasions, and the strength of each mechanism in a given context will determine whether native relatedness has an overall positive or negative effect for exotic species (Fridley et al., 2007;Tomasetto et al., 2019).
The enemy release hypothesis states that exotic plant species invading new areas may escape natural enemies and have lower herbivore pressure (Colautti & MacIsaac, 2004;Keane & Crawley, 2002).
This may enable exotic plants to shift resource allocation from defense to growth and reproduction, giving the invading species a competitive advantage over extant natives (Blossey & Notzold, 1995).
The strength of the enemy release effect in invasion is also influenced by the degree of phylogenetic relatedness between invading exotic species and the native community because related species are more likely to share specialist herbivores due to phylogenetic conservatism of insect host choice (Dawson et al., 2009;Lockwood et al., 2001). In species-rich tropical forests, an invading species will likely have a confamiliar or congener in the native community (Gentry, 1988;Turner, 2001), thereby overcoming the "habitat filtering barrier," but being faced with a potential "biotic resistance barrier" which may limit the invasiveness of the exotic species. However, an invading species may have varied responses in differing contexts, depending on the "receiving" community, including the presence and abundance of native confamiliars.
The empirical support for these hypotheses is mixed (Felker-Quinn et al., 2013;Jeschke et al., 2012;Liu & Stiling, 2006;Meijer et al., 2016), and studies are lacking in tropical regions. There is some support for the enemy release hypothesis in tropical forests (Dawson et al., 2009;DeWalt et al., 2004;Liu et al., 2007), including C. hirta, which is highly invasive in Hawai'i where its success is partially due to escape from natural enemies that limit populations to forest edge habitats in its native range (DeWalt et al., 2004). In Tanzanian forests, herbivory rates across multiple exotic species were found to be higher when there were more related species present in the native community (Dawson et al., 2009), implying a role for natural enemies in the invasion process.
However, there is little consensus within tropical forests on the role of native communities, with positive, negative, and nonsignificant relationships observed between exotic and native components (Döbert et al., 2017a;Murphy et al., 2008;Teo et al., 2003). This conflicting empirical evidence for these hypotheses is due in part to the lack of studies within tropical forests (Chong et al., 2021) and the context dependency of invasion mechanisms, with the sign and magnitude of any relationship varying with the observed conditions (Catford et al., 2021;Jeschke et al., 2012;Pyšek et al., 2020). In addition, it is unclear how these mechanisms might be affected locally by different abiotic factors, such as the level of disturbance, which is known to alter plant communities (Arellano-Rivas et al., 2016;Laurance et al., 2002Laurance et al., , 2011 and in turn their herbivore communities (Basset et al., 2012;Novotny et al., 2006). Therefore, exotic invasions and the hypotheses explaining their success (i.e., biotic resistance or enemy release) need to be studied along environmental and disturbance gradients.
The highly invasive tropical shrub C. hirta (L.) D.Don (Melastomataceae) is native to Central and South America (Mexico to Paraguay, and the Caribbean) but is invasive across tropical Africa, Asia, Australia, and Pacific Islands (CABI, 2020), including being the only widespread species invading intact forests in Sabah, Malaysian Borneo (Waddell, Chapman, et al., 2020). The success of C. hirta in Hawai'i may be due to escape from natural enemies (DeWalt et al., 2004) because there are no native members of the Melastomataceae there. Host-sharing by herbivores may be more likely in other parts of C. hirta's introduced range where it co-occurs with confamiliars, such as Borneo (DeWalt et al., 2004). Due to C. hirta's ability to invade and subsequently modify communities of tropical forest interiors (DeWalt et al., 2004;Waddell, Chapman, et al., 2020), it is important to improve understanding of the mechanisms affecting invasion by this species in tropical forests. In this study, we examine evidence for the roles of natural enemy release and biotic resistance in invasion of C. hirta within agricultural landscapes of Sabah, Malaysian Borneo, where C. hirta co-occurs with native species of Melastomataceae. First, we look for evidence consistent with enemy release by testing if herbivory in C. hirta is lower compared to native Melastomataceae (Melastoma spp.) shrubs that occupy a similar ecological niche. Second, we look for context dependency of enemy release by testing whether herbivory rates on C. hirta are influenced by proximity to, and herbivory rates of, native Melastoma plants (i.e., evidence of biotic resistance), and the influence of local canopy cover. Finally, we test whether herbivory reduces the reproductive output of C. hirta plants, and therefore might play an important role in mitigating population growth and spread.

| Related native Melastoma plants on Borneo
Related (confamilial) species belonging to the Melastoma genus were chosen as the native comparison. There are 41 native Melastoma species found in Borneo (Wong, 2016), and identification in the field (as in this study) is difficult due to very similar morphology and hybridization (Wu et al., 2019). However, only three species have distributions within our sampling area (M. malabathricum, M. beccarianum, and M. ultramaficum), with the vast majority of plants encountered being the widespread and very common Melastoma malabathricum (Wong, 2016). M. malabathricum is a non-forest specialist shrub that occupies similar habitats to C. hirta (i.e., roadsides, open-structure forest), and it flowers all year round, as does C. hirta. Most Melastoma individuals would have been M. malabathricum, but because we did not carry out detailed observations of flower morphology or genetic studies, we cannot be certain, and so we refer to Melastoma spp. throughout.

| Study sites
We carried out surveys of herbivory on C. hirta and native Melastoma Forest sites were conservation set-asides (hereafter "forest sites"), with most specified as being of High Conservation Value (HCVs, Senior et al., 2015), as part of the certification process for sustainability (Roundtable on Sustainable Palm Oil [RSPO], 2018). Forest sites had been commercially logged with varying intensity prior to plantation development, likely between ~1991 and 2009 (Reynolds et al., 2011), and were probably originally retained because they were unsuitable for oil palm cultivation (e.g., too steep or rocky).
Forest sites differed in terms of canopy cover and vegetation structure, reflecting these different levels of disturbance prior to plantation development, and varied in size from 3.25 to 1375 ha (forest area surrounded by plantations), with some sites (n = 6) adjoining large tracts of forest beyond plantation boundaries (Table S1). Sites were >1 km apart to avoid spatial autocorrelation.
At each site, 100-m transects were established in two forest habitats, one at the forest-plantation edge (hereafter "forest edge"; Waddell, Chapman, et al., 2020) and one in the forest interior (on average 53 m from nearest forest edge; hereafter "forest interior"; Figure 1c). The forest interior transects were located in regenerating forest following clearance, either along logging skid trails, old roads, or abandoned oil palm terraces (Table S1). These two forest habitats were chosen so that our surveys spanned a gradient of canopy cover resulting from previous disturbance that might influence local densities of plants and their herbivore communities. Closed canopy intact forest was not surveyed due to the very low density of C. hirta in this habitat (Waddell, Chapman, et al., 2020). The two forest habitats we surveyed differed in terms of light levels and species present, including density of C.
In total, we obtained data for 39 transects (one forest interior  Table S2), in line with our leaf sampling strategy (see below).

| Herbivory measurements
In this paper, we use the term "herbivory" to refer to any type of leaf damage including by invertebrates (e.g., galls, tissue removal by chewing and leaf scarring), mammals, and pathogens, reflecting the difficulty in distinguishing causes of leaf damage (Coley & Barone, 1996). For each individual plant, both C. hirta and Melastoma sp., we recorded two measures of herbivory to capture different sources of variation that may occur because of herbivore feeding behavior: herbivory occurrence and intensity ( Figure 1d). Herbivory occurrence was recorded by counting the total number of leaves and the number that had sustained herbivore damage to derive a proportion of damaged leaves per plant. To measure herbivory intensity, the first four fully expanded leaves closest to the shoot apical meristems were sampled. New leaves were chosen to minimize effects of leaf age on accumulated herbivory over time. Both sides of the leaves were scanned into high-resolution color images within 6 hours (to avoid leaf degradation) using an Epson Perfection V39 Scanner. Herbivory intensity for each leaf was graded by eye from scanned images to the nearest 10%, with the lowest category of <10% further divided into 0%, 0-2%, 2-5%, and 5-10%, due to the relatively high proportion of leaves within this category of herbivory.
The middle value for each category was used as the herbivory intensity score in analyses (e.g., 0, 1, 3.5, 7.5, 15%) and treated as a continuous variable for analysis (see "Statistical analyses" section).
We used "by eye" measurements carried out by the same person (EHW) to reduce any observer biases. This robust measure is a costeffective approach for measuring large datasets in comparative studies and has been shown to yield similar results to digital image processing methods (Kozlov & Zvereva, 2018 and references within).
We used it to integrate damage from multiple different sources (e.g., pathogen attack as well as removal of tissue by herbivorous insects) and it has the benefit that it does not require the original leaf margin to be manually "re-created" as would be necessary for imagining software (e.g., ImageJ; Johnson et al., 2016).
To test whether leaf herbivory reduced reproductive fitness of C. hirta, the numbers of reproductive organs (buds, flowers, immature fruits, and ripe fruits) per plant were recorded. These were also recorded for native Melastoma spp. plants to draw comparisons between the species groups.

| Variables influencing herbivory
To examine their influence on herbivory damage in C. hirta, we recorded the distance to the closest native Melastoma plant and the herbivory rate on this plant. In addition, we recorded habitat type (forest edge vs. forest interior) and local canopy cover, as light levels influence herbivory rates in some species (Eichhorn et al., 2006;Folgarait et al., 1995;Whitham, 1991). We also accounted for C. hirta plant size and the density of surrounding C. hirta plants, given that adult plants can act as reservoirs for enemies to attack nearby conspecific saplings and seedlings (the Janzen-Connell hypothesis of negative density dependence (Connell, 1971;Janzen, 1970).
To test whether there is spillover of herbivores and pathogens from native Melastoma plant onto C. hirta plants, we calculated a composite variable, termed "spillover index," to capture simultaneously the proximity to the nearest native Melastoma spp. plant, and the level of herbivory on that Melastoma plant. We hypothesized that these would jointly drive spillover. This composite variable uses an exponential distance decay function (Moilanen & Nieminen, 2002), which is commonly used in landscape ecology (Chandler & Hepinstall-Cymerman, 2016;Lowe et al., 2022) and assumes a constant decline in spillover rate with increasing distance. This distance decay is multiplied by herbivory damage on that Melastoma plant, such that a Melastoma plant at the same distance will have a higher spillover index value if it is more damaged. The equation we formu- where h is herbivory damage on the nearest native Melastoma plant, d is the distance between C. hirta and that nearest Melastoma plant, and x is the distance decay parameter.
We investigated different values of x and found the exact choice of x in our spillover index is not crucial for our conclusions. Nevertheless, we choose the value of x based on model AIC, which was x = 2 for C. hirta herbivory occurrence and x = 5 for C. hirta herbivory intensity (see Appendix S1 and Table S3 in the Supporting Information for details of spillover index calculations and parameter optimization).
The degree of forest disturbance was captured by surveying the two habitats, either highly disturbed forest edges or previously disturbed forest interiors. Percentage canopy cover was recorded every 20 m along transects (five points per transect, with the mean of four measurements facing each cardinal direction per point; recorded by the same observer each time) using a concave spherical densiometer (www.fores try-suppl iers.com). As an indicator of the light levels experienced by each surveyed plant, we used the nearest canopy cover measurement or the average of the two closest if the plant was equally spaced between two densiometer measurements.
We recorded the total number of leaves as a measure of C. hirta plant size. We recorded the presence/absence of C. hirta in each 1 m section (and 1 m either side of the transect line; 2 m 2 ), along the length of the 100-m transect (total sample area 200 m 2 ) and defined the "local density of C. hirta" as the number of plants within 5 m either side of each focal C. hirta plant (presence out of ten 2 m 2 sections; 1-10).

| Statistical analyses
We modeled herbivory occurrence and herbivory intensity separately, as we found these two herbivory measurements to be only weakly positively correlated (Spearman rank correlation coefficient = 0.28, p < .001; SI Figure S1), and they measure different components of variation in herbivory. We used generalized linear mixed models (GLMMs) using the "lme4" R package (Bates et al., 2015). For "herbivory occurrence," all models had a binomial error distribution with the frequencies of damaged and undamaged leaves as the response variable.
"Herbivory intensity" was natural log +1 transformed and modeled with a Gaussian error distribution. While the fixed effects of models varied according to the question being addressed (detailed below), the random effect structure stayed the same for all models (individual plant nested within transect nested within site). To improve model convergence across all models, continuous fixed effects were scaled to unit variance and centered on zero. For each model, all possible combinations of terms in the maximal model were compared using the "dredge" function in the "MuMIn" R package (Bartoń, 2013) and the best model selected based on minimizing the Akaike information criterion (AIC c ). The variance explained by the best model (conditional R 2 ) and each of the fixed effect terms (marginal R 2 ) were estimated for each model using the "rsquared.GLMM" function in "MuMIn" (Nakagawa et al., 2017). The variance explained by each fixed effect included in the top model was estimated separately by removing effects one by one and calculating the difference in the marginal R 2 value.
To test whether herbivory damage was lower in exotic C. hirta than in related native plants, herbivory was compared between species (Table 1: Models 1a and 1b). GLMMs were fitted for herbivory occurrence and herbivory intensity separately, with fixed effects of species type (exotic or native), and an interaction with habitat (forest edge or forest interior) to test whether interspecific differences in herbivory varied with forest disturbance.
To test the drivers of herbivory levels in C. hirta, we fitted models which include the fixed effects of proximity to herbivory in native species (i.e., our spillover index), habitat (edge vs. interior), and local canopy cover (%), while controlling for plant size (total number of leaves per plant) and local density of C. hirta (number (from 0 to 10) of 2 m 2 transect sections with C. hirta present surrounding focal C. hirta). These two models (Table 1: Models 2a and 2b) included those C. hirta plants that were co-occurring with a Melastoma individual only (i.e., within 10 m; n = 196 individuals). To test whether these variables influence herbivory differently under different levels of disturbance (i.e., forest edge vs forest interior), an interaction with habitat type was included for all fixed effects. We re-fit these two models (2a and 2b) using our full dataset (all C. hirta plants recorded, n = 366, not just those co-occurring with native Melastoma), to test the effect of habitat, local canopy cover, plant size, and C. hirta density on herbivory in C. hirta, but without testing for the spillover index (Table 1: Models 3a and 3b).
To examine the effect of herbivory on the reproductive output of C. hirta, a GLMM was fitted with the total number of reproductive organs per plant (buds, flowers, immature fruit, and ripe fruit) as the response variable with a Poisson error distribution and fixed effects of herbivory occurrence, herbivory intensity, habitat type, local canopy cover, plant size, and local density of C. hirta (Table 1: Model 4).
We also tested for interacting effects between all fixed effect terms and habitat to examine whether these variables influence reproductive output differently under different disturbance levels.

| Comparison of herbivory in C. hirta and related natives
Herbivory was significantly lower in exotic C. hirta than in native Melastoma plants for both herbivory occurrence (15.0% fewer leaves damaged in C. hirta plants compared with Melastoma plants; Table 1 and Figure 2a; p < .001) and herbivory intensity (15.9% lower per leaf for C. hirta compared with Melastoma; Table 1 and Figure 2b; p < .001).
For both measures of herbivory (occurrence and intensity), we found higher damage on C. hirta plants closer to related native plants with high levels of herbivory (i.e., higher spillover index; Figure 3a and Table 1; p < .001). Disentangling the two components of the spillover index showed that the rate of herbivory damage in native Melastoma was more associated with herbivory in C. hirta than the distance between the plants (Appendix S1; Table S4). For example, there was no difference in expected herbivory occurrence in C. hirta plants that were 0-10 m from a native Melastoma plant with low (1%) herbivory intensity but expected herbivory occurrence increased to 13% in C.
hirta that were between 0 and 10 m from a native plant with the highest (92.5%) recorded native herbivory damage (Appendix S1; Table S4).

| Influence of canopy cover and habitat on exotic herbivory
Canopy cover was nearly twice as high inside the forest than at the forest edge (median cover on edge transect = 49%, range = <1-94%, interior = 83%, range = 8-96%; Figure S2; p < .01). We found that herbivory occurrence in C. hirta was significantly higher at locations with higher canopy cover (Figure 3b and Table 1; p < .05), although there was no effect of habitat type on herbivory (occurrence and intensity) for either exotic or native species, despite differences in canopy cover between forest edge and forest interiors (Figure 2c).
This suggests that light levels are influential in determining rates of herbivory in C. hirta, although canopy cover explained less variation than proximity to native plants in our models (Table 1). On average, there was little difference in plant size between habitats (Table S2; p = .38) or between species (Table S2; p = .22). While we found no direct influence of habitat type on herbivory, there was a significant interaction between plant size and habitat (Table 1; p < .05) which revealed that the occurrence of herbivory was lower on larger C. hirta plants, but only within the forest interior. There was little difference in the local density of C. hirta at forest edges (median 8 plants surrounding focal C. hirta plant) compared with interior transects (6 plants; Table S2; p = .49), and we found no evidence of densitydependent herbivory occurring on C. hirta across models (i.e., there was no significant effect of local C. hirta density on either herbivory measurement; Table 1).
We found evidence that canopy cover, plant size, and proximity to native Melastoma affect herbivory in C. hirta but overall, we found considerable unexplained variation around the significant relationships between herbivory in C. hirta and our tested drivers. The data TA B L E 1 Summary of best generalized linear mixed models for herbivory in exotic shrub Clidemia hirta, in Sabah, Malaysian Borneo. Model numbers correspond to our three hypotheses (see aims) and full model descriptions are outlined in "Statistical analyses" in the methods, including all fixed effects tested. "Herbivory occurrence" = percentage of leaves per plant that are damaged by herbivory and "Herbivory intensity" = percentage of leaf area that is damaged by herbivory. R 2 m = Marginal R 2 , which indicates the R 2 for all fixed effects included in the model, R 2 c = Conditional R 2 , which indicates the R 2 of the full model (Fixed effects + random effects). Interactions between fixed effects are denoted by an "x." shown in the scatterplots (Figures 3a, b), and the low R 2 terms for fixed effects in our models (R 2 ranging 0.004-0.231; Table 1) suggest there are other factors influencing herbivory in C. hirta that we did not record.

| Effect of herbivory on reproduction
We found a significant negative relationship between herbivory intensity and the number of reproductive organs produced by C. hirta ( and more organs were found on large plants (Table S2; p < .001).

| DISCUSS ION
We found evidence that herbivory damage in invasive C. hirta is affected by proximity to native Melastoma plants with high rates of herbivory. This suggests that spillover of specialist herbivores and disease is occurring, and potentially impacting the performance of C. hirta within its introduced range in Sabah, Malaysian Borneo.
We found that herbivory was greater in areas with higher canopy cover, showing that C. hirta plants in more intact forests experience increased herbivory damage, offering a potential means to suppressing C. hirta populations in structurally intact forests. Higher rates of herbivory were associated with lower reproductive output of C. hirta plants, as measured by the number of reproductive organs present, suggesting that damage by herbivores may be having a negative impact on the fitness of C. hirta. Overall, we found C. hirta plants experienced lower herbivory and were more likely to be reproductively active compared with native Melastoma plants ( Figure S3), which is likely to place C. hirta at a competitive advantage over confamiliar native species. However, despite being ecologically very similar (i.e., shrubs, similar maximum heights, flowering year-round), there may be temporal differences in the reproductive output of C. hirta and Melastoma spp.; therefore, year-round monitoring would be needed to confirm our observations.

| Evidence for enemy release in C. hirta invasion in Sabah
Across our sites, we found C. hirta to have much lower herbivory damage than native Melastoma spp. (6.1% cf 20% herbivory intensity per leaf), providing potential evidence for enemy release occurring in Sabah. However, the overall levels of herbivory intensity (i.e., percentage leaf area removed) we recorded in C. hirta were higher F I G U R E 2 Herbivory rates in exotic Clidemia hirta and native Melastoma spp. plants at forest-oil palm edge (light green shading) and within disturbed forest set asides (dark green shading), in Sabah, Malaysian Borneo. (a) Difference in herbivory occurrence between species and habitat ("herbivory occurrence" = percentage of leaves per plant that are damaged by herbivory). (b) Difference in herbivory intensity between species and habitat ("Herbivory intensity" = percentage of leaf area that is damaged by herbivory). Boxplots show the median (central line) and mean (yellow diamond) values. Significant p-values displayed are from ANOVA on generalized linear mixed models (***p < 0.001). Details of tests in Table 1 (Models 1a and 1b). to the argument that host-sharing is occurring in our sites, and hence some evidence of biotic resistance.

| Evidence for biotic resistance to C. hirta invasion by herbivores and pathogens
Our finding of a positive association between herbivory damage in C. hirta and proximity to native related species may mean there are shared diseases and herbivores that are adapted to feeding on Melastomataceae plants, including C. hirta, despite not being native to Sabah. This finding agrees with other studies that found evidence of phylogenetic conservatism of herbivores in tropical plant species (Ødegaard et al., 2005;Weiblen et al., 2006), including in exotic species (Dawson et al., 2009). In our study, herbivores and pathogens were not explicitly observed but evidence of the same family of moth (tussock moths, Lymantriidae; Georg Hantke pers. comm.) was observed on leaves of both exotic and native species (Figure S4a-b) as well as similar patterns of disease ( Figure S4c-e). The significance of the spillover index indicates that melastome herbivores and pathogens are driving patterns in herbivory that we observe, but further investigation to identify the role of specific herbivores and pathogens would confirm whether they belong to specialist or generalist groups.
The negative association we found between herbivory intensity and number of reproductive organs suggests that damage by herbivores and pathogens may reduce C. hirta fitness within Sabah. However, despite our significant finding, our effect sizes are small, which tells us other factors are much more influential in driving reproduction in C. hirta in Sabah (e.g., plant age, nutrients, pollination, that were not measured in the study). Nevertheless, herbivory can reduce fitness as plants move resources away from reproduction to defense (Eichhorn et al., 2010;Marquis, 1984), and the patterns we observed could potentially affect local population growth and dispersal, reducing spread rates and local dominance of C. hirta in Sabah. However, C. hirta is widespread in these landscapes, despite increased herbivory pressure in sites with higher canopy cover and reduced reproductive output, including in continuous "old-growth" rainforests (with ~92% mean canopy cover; . Although herbivory intensity is apparently not high enough to exclude C. hirta from these forests, it may be enough to regulate populations, preventing C.
hirta from gaining the competitive advantage it has in Hawai'i to become a dominant species in sites with intact forest understory.

F I G U R E 3
Relationship between herbivory occurrence in exotic Clidemia hirta plants and (a) index of spillover from native Melastoma spp.
(n = 196) and (b) local canopy cover (n = 366). "Spillover index" = composite variable based on distance decay, combining distance to nearest native plant and herbivory intensity recorded on that plant (Appendix S3 for full details). Regression lines with a 95% confidence interval (gray shading) are on the raw data and does not control for other variables included in the GLMM. Significant p-values displayed are from ANOVA on generalized linear mixed models. Details of tests in Table 1.
This effect of herbivory would be a form of biotic resistance called biotic containment (Levine et al., 2004), in which natural enemies contribute to limiting populations by lowering growth, reproduction and spread. However, a thorough investigation into the impact of herbivory on growth rates and reproductive output over time using natural enemy exclusion experiments such as in DeWalt et al. (2004) is required to draw robust conclusions on the contribution of herbivory on C. hirta fitness, population dynamics and persistence in Sabah.

| Role of canopy cover in herbivory pressure
We found the occurrence of herbivory on C. hirta and native  (Poorter, 1999;Steinger et al., 2003). This adaptation may explain the relationship we found between canopy cover and C. hirta herbivory, if thinner leaves are less protected from herbivory, and/or have higher nutritional value. Indeed, SLA data collected for C. hirta plants (n = 294) in the same forest study sites showed strong increasing trends for thinner leaves over a gradient from oil palm plantation to intact forest, with increasing canopy cover ( Figure S3). This trend (i.e., higher SLA in low light conditions) has also been observed in C. hirta in a common garden experiment comparing native Costa Rica and introduction Hawai'ian populations (DeWalt et al., 2004). Support for a positive relationship between herbivory and SLA in tropical forests is mixed, with support for (Coley, 1983;Coley, 1988) and against (Poorter et al., 2004), and so evidence in our study for relationships between SLA and canopy cover requires further investigation to explore influences on herbivory patterns. Also, greater herbivory in areas with high canopy cover may be explained by differences in the invertebrate and microbial communities driven by abiotic conditions (e.g., temperature and humidity; Boyle et al., 2021) or biotic factors (e.g., understory pant community; Novotny et al., 2006;Basset et al., 2012), and warrants further investigation.

| Other potential drivers
This study tested a small part of the overall system in which interactions between exotic C. hirta and native Melastoma spp. are occurring, and although significant, our fixed effects of spillover and canopy cover in our models explain low variation (R 2 values in Table 1

| Conclusion
C. hirta is one of the most prevalent exotic invaders of tropical forests, including forest patches embedded in oil palm landscapes in South-East Asia, which are crucial for maintaining biodiversity within human-modified tropical landscapes. Our findings that there is reduced herbivory and potentially higher reproductive output in more open habitats, may result in the spread and colonization of C. hirta in the most degraded of these forest patches.
Overall, our results suggest that herbivory pressure for exotic species is higher when native confamiliars are present in the native flora, and shared herbivores may act as a "biological control," limiting their reproduction and thus the long-term spread. These results highlight that other exotic species that have no native confamiliars may pose the highest risk to these human-modified habitats, especially those species with the traits known to increase colonization within tropical forests (Dawson et al., 2009;Waddell, Chapman, et al., 2020). By restoring these forests to have intact canopy cover, the number of exotic species will decrease (Waddell, Chapman, et al., 2020) and this high canopy cover may also limit the spread of C. hirta due to reservoirs of herbivores inside intact forest that provide a natural control, lowering reproductive outputs and therefore fitness within Sabah.

AUTH O R CO NTR I B UTI O N S
EHW, LFB, MH, JKH, and DSC conceived the ideas. EHW, LFB, MH, JT, JKH, ABS, and DSC designed the methodology; EHW and ABS collected the data; EHW and DSC analyzed the data; EHW led the writing of the manuscript.

CO N FLI C T O F I NTER E S T S TATEM ENT
The corresponding author confirms on behalf of all authors that there have been no involvements that might raise the question of bias in the work reported or in the conclusions, implications, or opinions stated.

DATA AVA I L A B I L I T Y S TAT E M E N T
The data that support the findings of this study are openly available at NERC EDS Environmental Information Data Centre. https://doi. org/10.5285/2e2e2 970-eb78-4005-b271-7101e 615186a (Waddell et al., 2023).