Release from aboveground enemies increases seedling survival in grasslands

1. Plant enemies can influence plant community assembly and structure. However, it is unclear how insect herbivores and fungal pathogens affect seedling recruitment. Complex interactions with competition and resource availability make it difficult to isolate the effect of enemies. This uncertainty can impede understanding of community assembly drivers, species coexistence and trophic interactions; and limits hypothesis testing such as of the enemy release hypothesis, a key hypothesis in invasion biology. Using a novel species-specific approach, we examine how enemies affect seedling survival and recruitment of 16 grassland species. 2. We planted seedlings of 16 native species from two functional groups (C4 grasses and non-legume forbs) into two grassland sites (early and mid-succession). We hand-painted 1512 individual seedlings with pesticides (insecticide and fungicide) over the course of one growing season to enforce aboveground species-specific release from enemies and tested whether it enhanced survival relative to untreated controls. Applying treatments to


| INTRODUC TI ON
Enemies such as herbivores and pathogens play an important role in the assembly and maintenance of plant communities (Grunberg et al., 2023;Mordecai, 2011).The effect of enemies on plant recruitment is likely to be particularly influential: recruitment represents a key assembly step linking dispersal from the wider species pool to establishment in local communities (Thompson et al., 2020) and juvenile plants are especially susceptible to enemies (Bruns et al., 2022;Zhao et al., 2020).Early seedling survival is thus an important performance metric determining recruitment success (Bohl Stricker & Stiling, 2013;Zhang et al., 2023).However, explicit measures of seedling survival are often overlooked (Brian & Catford, 2023;Lonardi et al., 2023).When seedling survival has been studied in the context of enemy effects, the focus has largely been on mammalian herbivores (e.g.Borer et al., 2014;Furey & Tilman, 2023;Lonardi et al., 2023) despite evidence that insect herbivores and fungal pathogens play prominent roles in determining community structure (Seabloom et al., 2017).
A range of community-and individual-level plant characteristics may interact with insect and fungal pathogens to affect seedling survival, creating context dependence and complicating predictions of how enemies affect plant recruitment (Catford et al., 2022).A community that is phylogenetically clustered may share many specialists and experience high enemy regulation, while phylogenetically diverse communities could experience dilution of enemy pressure (Leonard et al., 2023).Abiotic factors such as resource levels affect plant investment into growth and defence, and thus the overall impact that enemies have (Hahn et al., 2021).Within communities, different functional groups experience pressure from different enemies.Evidence from a globally distributed experiment suggests, for example, that the dominant enemy group of C4 grasses is fungal pathogens, while non-legume forbs are targeted more by insect herbivores (Ebeling et al., 2022).Release from enemies also appears to benefit non-legume forb biomass to a much greater extent than C4 grass biomass, a response that also depends on broader community diversity (Seabloom et al., 2018).Within functional groups, interspecific trait variation alters plant responses to enemies.For example, species that invest heavily in growth at the expense of defence are more likely to suffer high levels of damage (Blumenthal et al., 2009).
Despite the wide variation in effects that enemies have on plant performance across communities and species, many studies use greenhouse experiments or focus on one or two plant species (e.g.Dewalt et al., 2004;Eckberg et al., 2014;Louda & Potvin, 1995).This limits the scope of potential inferences on how enemies affect recruitment into plant communities (Ojha et al., 2022).
The effect that insect herbivores and fungal pathogens have on communities is often tested by applying pesticides at a plot level (e.g.Grunberg et al., 2023;Heckman, 2023;Heckman et al., 2017;Zhao et al., 2023).Plot-level application is an effective tool to determine the impact of enemies on total biomass production, diversity or turnover in plant communities (Grunberg et al., 2023;Heckman et al., 2017).However, as well as impacting plant diversity, effects of plot-level enemy removal are themselves affected by the starting plant diversity of the plot (Seabloom et al., 2017).Because plant diversity also affects recruitment success of new seedlings (Catford et al., 2020), plot-level pesticide applications cannot disentangle enemy effects from community diversity effects on recruitment success of individual species (Agrawal & Maron, 2022).A more direct way to test for enemy effects on recruitment is to apply pesticides to individual seedlings in natural communities.This strategy facilitates direct observation of the effects of release on target individuals while being able to account for community context.We are aware of very few studies that apply pesticides to individuals to directly link enemy effects to plant performance in natural communities (but see Dewalt et al., 2004;Morris et al., 2022;Vasquez & Meyer, 2011).
Appropriately testing the effect of enemies on recruitment is also important for understanding the enemy release hypothesis (ERH), the best-known hypothesis explaining biological invasions (Enders et al., 2018).The ERH is also a key mechanism by which enemies can affect general plant community dynamics (Mordecai, 2011).The ERH states that, on movement to a new range, exotic species may escape the regulatory effects of enemies from their home range and can then become competitively dominant over native species in their invaded range (Keane & Crawley, 2002).However, the link between loss of enemies and increased performance remains understudied (Agrawal & Maron, 2022;Brian & Catford, 2023;Prior et al., 2015), especially early in invasion (Hawkes, 2007).Therefore, it is difficult to reliably link enemy release to the demographic advantages that are often exhibited by exotics.A recent review of ERH studies showed that only 31% of studies measure release from enemies and link this to exotic performance, and only 6% of studies consider exotic survival in the context of the ERH (Brian & Catford, 2023).
Studies of exotic performance are often confounded by different environmental conditions in the home and invaded ranges, which interact with enemy diversity or impact and make it difficult to isolate the effect of enemies on plant performance (Brian & Catford, 2023; results also support the key mechanism (increased performance following release from enemies) underlying the enemy release hypothesis.Enemy release may therefore aid initial recruitment of plants during the invasion process.

K E Y W O R D S
Cedar Creek Ecosystem Science Reserve, enemy release hypothesis, exotic plant invasion, grassland, pathogen, pesticide, plant-herbivore interactions, traits Xu et al., 2023).Surveys or experiments on naturally occurring exotic species are often biased towards conditions that facilitate exotic establishment and success (such as disturbed or high-resource locations), providing misleading estimates of exotic performance (Pearson et al., 2022) and producing context dependence (Catford et al., 2022).However, applying pesticides to individual native seedlings and comparing performance against non-treated individuals tests the mechanism of enemy release: it releases specific individuals from enemy regulation, while co-occurring individuals in the community are still faced with enemies.Evidence for increased performance in released individuals relative to non-released individuals would then mechanistically support the link between ecological release from local enemies and performance.We emphasise that this approach using native seedlings does not test the ERH: it cannot determine whether enemy release has contributed to the success of a given invasive species, or the prevalence of enemy release among invasive species, nor does it account for the fact that exotic species may systematically differ in traits to native species (van Kleunen et al., 2010).However, it causally demonstrates that release from enemies can benefit individual recruitment, a link that has not been fully investigated in the ERH literature (Brian & Catford, 2023).
We carried out a grassland field experiment to test the impact of enemies on seedling recruitment over one growing season.We studied 16 native plant species from two functional groups (C4 grasses and non-legume forbs) released from two enemy guilds (aboveground fungal pathogens and aboveground insect herbivores) and examined the effects of release in communities at two different successional stages.
We applied pesticides by hand-painting individual target plants, rather than spraying the whole community in a target plot.We ask three questions: (1) Does release from insect herbivores or fungal pathogens increase seedling survival in natural communities, and is this effect consistent between C4 grasses and forbs?(2) How do the benefits of release vary with community-level abiotic and biotic characteristics (soil moisture, light availability, successional stage, species richness and phylogenetic relatedness to focal seedlings)?(3) Can variation in the benefits of release between species be explained by species-level characteristics?Our experiment tests enemy effects on recruitment into grassland communities and interrogates the contexts under which these effects are strongest, providing generalisable evidence for the role insect herbivores and fungal pathogens play in community assembly.By experimentally simulating the core mechanism of enemy release (an enemy-free plant recruiting into an enemy-regulated community), we also provide causal insight into how enemy release may facilitate exotic plant establishment and thus increase invasion potential.

| Study site and species
Our experiment took place at Cedar Creek Ecosystem Science Reserve (hereon Cedar Creek) in Minnesota, USA.Cedar Creek has nitrogen-limited sandy soils, annual precipitation of ~780 mm and mean annual temperature of 6.72°C (averaged from 1987 to 2016 at Cedar Creek weather station).We studied 16 species, eight C4 grasses and eight non-legume forbs (Table 1).We chose these functional groups as they have previously displayed the most divergent responses to enemy removal (Seabloom et al., 2018).The eight species within each group vary in aboveground traits (Figure S1).While the forbs only represent three families, these families represent the majority of forbs found in our study field.Our goal was to use experimental species that naturally occur in those sorts of communities but were absent (or very rare) from our sites within the field-the 16 chosen species represent the maximal variation that could be achieved using those criteria.All 16 species are native to Minnesota and present at Cedar Creek.
We sourced seeds from two native seed suppliers who collect seeds from local grassland populations (Prairie Moon, prair iemoon.com; Prairie Restorations, prair ieres to.com).Individual seeds were planted in 12 cm deep 'cone-tainers', which were filled with sterilised soil (heated for 24 h at 80°C) taken from Cedar Creek with a thin layer of vermiculite added on top.These were grown in a glasshouse for 1 month and irrigated regularly, after which time the seeds of all 16 species had germinated.The glasshouse was completely enclosed and monitored regularly for pests, protecting seedlings from enemies throughout germination and initial growth.

| Experimental design
We established our experiment in the Lawrence Strips (45.414°N, −93.184°W), a single old field at Cedar Creek approximately 100 m wide and 650 m long.This field was incrementally abandoned from agriculture: beginning in 1974, a 100 m × 20 m strip was abandoned each year, moving north through the field with the final strip abandoned in 2008 (Inouye et al., 1994).The field therefore contains a gradient of plant communities at different successional stages.We established plots in two strips in 2022: 'early succession' (20 years old, abandoned 2002) and 'mid succession' (41 years old, abandoned 1981) (Figure S2).As both strips are in the same field separated by just 420 m, these two sites are similar in all key aspects (e.g.soil type, elevation, climate) apart from successional stage.Each set of plots was surrounded by a fence to exclude vertebrate herbivores.The diversity and composition of the two successional stages differed.
Four of the study species (A.millefolium, A. syriaca, A. tuberosa, S. scoparium) were present at very low levels in the plots prior to experiment establishment (all <1% mean abundance).The other 12 species were present elsewhere in the field or surrounding fields but absent from all experimental plots.
To confirm that our sites adequately represent their respective successional stages in the local region and that differences between our sets of plots are related to successional stage, we compared the taxonomic composition of our plots to plots in nearby old fields of comparable ages (see Catford et al., 2023).These comparisons showed clear differentiation between our two sites (Figure S3a), and our communities fell within the range of compositions observed for their respective successional stages (Figure S3b), indicating that our communities provide sound examples of their successional stages.This aligns with previous findings that old field succession largely follows consistent and predictable trends at Cedar Creek (Clark et al., 2019).We nevertheless emphasise that plots should be interpreted as independent replicates of our target sites, not independent replicates of all grasslands at our two successional stages (Colegrave & Ruxton, 2018).
On 15 June 2022, we planted 1512 seedlings (approximately 1 month old) across 288 unmanipulated plots, with all existing vegetation undisturbed and intact.Plots were 1 m × 1 m, separated by 0.5 m walkways.Seedlings were planted into the central 0.5 m 2 of each plot to limit possible edge effects (Figure S2).Each plot received up to six individuals of a single species (some species had fewer than six individuals per plot due to low germination success, Table 1).
We had three experimental enemy removal treatments (described below), and each species had three replicate plots per treatment.
We therefore had 144 plots per successional stage (16 species × 3 treatments × 3 replicates), with species and treatments randomly assigned to plots to avoid spatial biases (Figure S2).Before planting, we measured the height of all seedlings.Seedlings were watered immediately after planting and as required throughout the first 3 weeks of the experiment to avoid excess mortality from drought stress.
The planted seedlings in each plot were assigned one of three enemy removal treatments: insecticide, insecticide+fungicide, or control.Pesticides were hand-painted on individual seedlings (i.e.only the seedlings were subject to enemy removal, not the whole community), using a foam brush and/or fine paintbrush until the leaves were visibly wet but not dripping.We did not do a fungicide-only treatment due to constraints on the overall area of the experiment and the time-intensive nature of treatment application; the effect of fungicide is interpreted as the difference between the insecticide and insecticide+fungicide treatments.We prioritised the insecticide single treatment as insect herbivory has generally been better studied (e.g.Turcotte et al., 2014) TA B L E 1 Study species in the experiment.
fungicide was Quilt (Syngenta Crop Protection, Inc., Greensboro, NC, USA; 7.5% Azoxystrobin and 12.5% Propiconazole).These pesticides have been shown not to affect plant growth (Seabloom et al., 2017(Seabloom et al., , 2018)).To control for the effect of painting, we painted water on plants in the control treatment.The treatments were applied every 2 weeks until the end of the growing season ( 31August 2022), with the first application occurring the day before being planted in the field (14 June 2022; 7 applications across the season).At the height of the growing season (beginning of August), we visually estimated percent cover (to the nearest %) of all species in the central 0.5 m 2 of each plot, as well as bare soil, litter and disturbances (e.g.gopher mounds).There were 60 species (including our planted species) across all plots.We also visually estimated insect damage and fungal damage on our (surviving) focal seedlings to the nearest percent, taking the average of five haphazardly selected leaves on each plant.Tissue removal (chewing damage) from the margin or interior of leaves as well as leaf miner tracks were counted as insect damage, while lesions, rust spots and mildew were counted as fungal damage.We used the training program and techniques described in We collected information on species phylogeny and functional traits to inform our statistical analyses.We built two phylogenetic trees using the V.PhyloMaker package (Jin & Qian, 2019).The first tree consisted of our planted 16 species, while the second consisted of all 60 species observed across all plots.We used the second tree to take a 'focal-species' approach (following Pinto- Ledezma et al., 2020), where we calculated the mean phylogenetic relatedness between our planted seedlings and all other species in the plot, weighted by the relative abundance of those species (abundance-weighted Mean Phylogenetic Distance, hereon 'phylogenetic distance').We collated data for five leaf traits [specific leaf area (SLA; mm 2 mg −1 ), leaf dry matter content (LDMC; mg g −1 ), leaf area (mm 2 ), leaf N (%) and leaf P (%)] for each of the 16 species.Leaf trait data were collected from neighbouring grasslands at Cedar Creek (Catford et al., 2019).As locally collected leaf N and P data only covered 15 and 12 of our 16 species respectively, we filled gaps first from a comparative North American grassland site (Craine et al., 2012) and then a global dataset (Reich & Oleksyn, 2004).We selected the five leaf traits to represent the broad axis of the plant fast-slow growth continuum (Craven et al., 2018).

| Statistical analysis
All statistical analyses were carried out in R v4.1.1 (R Core Team, 2021).See Supporting information Code S1.
2.4.1 | Questions 1 and 2: Does release from enemies increase seedling survival, is this effect consistent between C4 grasses and forbs, and how do community-level biotic and abiotic characteristics affect the benefits of release?
After confirming that pesticides reduced enemy damage (Figure S4), we ran a series of models to answer questions 1 and 2 (model set 1a, Methods S1).We used the proportion of seedlings surviving in a given plot as the response variable, and ran seven general linear mixed effects models (one for each observation date).Fixed effects were treatment (control, insecticide, insecticide+fungicide), successional stage (early, mid), functional group (C4 grass, forb), plot richness, phylogenetic distance, moisture and light availability, with individual species nested within functional group as a random effect.
We tested all two-way interactions: none were statistically significant (α = 0.05).We therefore present results from models that only contain main effects.Statistical significance of a variable is determined by a χ 2 -based likelihood ratio test comparing models with and without the factor of interest.Diagnostics for these models (and the ones described below) models were inspected using the DHARMa package (Hartig, 2022) and found to be acceptable.
We ran models for each of the seven observation dates separately (following Charles et al., 2018), as the dates are not independent and so date cannot be included as a factor (if a seedling is dead at time t, it is guaranteed to be dead at t + 1).This approach also allowed us to explicitly test the strength of the treatment effect at each time period.Given this multiple testing approach, we draw conclusions at a conservative Bonferroni-corrected α = 0.007 (0.05/7).However, we also explicitly tested for a treatment × time interaction by using the number of new dead seedlings per plot as As differences in the initial size of seedlings may have affected seedling survival, we ran a second set of seven mixed effects models to answer question 1 (model set 2, Methods S1).These models investigated individual seedling survival, with binary survival (yes/ no) as the response variable.Models used the binomial family (logit link) and had initial height, functional group, treatment and successional stage as fixed effects, with plot nested within species and functional group as a random effect to account for the plot-level variables of phylogenetic distance, moisture, light and plot richness (these could not be directly included in the model as fixed effects due to severe multicollinearity, as they are identical for all seedlings in a given plot).Initial height was measured when seedlings were transplanted from the glasshouse to the field experiment.We used the package glmmTMB (Brooks et al., 2017) for model sets 1 and 2.
We also confirmed that there were no differences in average seedling height between the treatments at the start of the experiment (F 2,1506 = 2.39, p = 0.092).

| Question 3: Can variation in the benefits of release be explained by species-level characteristics?
To answer question 3, we first ran individual-level models for each species independently (16 models total), to quantify the effect of the treatment on survival at the species level in the final time period (model set 3, Methods S1).Models used the binomial family (logit link) and had initial height, treatment and successional stage as fixed effects, with plot as a random effect to account for phylogenetic distance, moisture, light and richness.We observed pronounced heterogeneity in the benefits of release from enemies for individual species.We then used species effect sizes (where 'effect size' refers to the benefit a species received from enemy removal treatments relative to controls) as the response variable for three further analyses, testing whether variation among species effect sizes could be explained by phylogeny, species traits or damage rates.Each analysis was carried out twice (once for the 'insecticide' effect size, and once for the 'insecticide+fungicide' effect size).
Using the phylogenetic tree of our 16 planted species, we tested whether more closely related species had more similar effect sizes.
We used the phylosignal package (Keck et al., 2016) to test the null hypothesis that effect sizes were randomly distributed in the phylogeny.The phylogenetic signal was measured using the C mean index, and the null hypothesis was rejected if p < 0.05 (Keck et al., 2016).
We then carried out a PCA on leaf traits and extracted the first two axes (Figure S1).We examined the relationship between these PCA axes and effect sizes for each species using linear models (effect sizes logged to meet assumptions), with the PCA axes and functional group (grass or forb) as independent variables.We also compared effect sizes against the single traits of SLA and leaf N, as two of the most important plant traits determining rates of individual-and community-level enemy damage (Cappelli et al., 2020).Finally, to test whether species that received more damage in controls had larger treatment effect sizes, we compared insect and fungal damage on our control plants with effect sizes using a linear model (effect sizes logged to meet assumptions) with damage rate and functional group as independent variables.For the above three sets of analyses, two species were outliers based on Cook's distance (M.fistulosa for the insecticide+fungicide treatment and P. virgatum for the insecticide treatment).We repeated analyses with these outliers excluded, which did not change overall results.Interactions between functional group and the other variables of interest were always found to be non-significant; we therefore present the main effects only.

| RE SULTS
3.1 | Question 1: Does release from enemies increase seedling survival in natural communities, and is this effect consistent between C4 grasses and forbs?
Pesticide application significantly reduced insect damage in the insecticide treatment, and reduced both insect and fungal damage in the insecticide+fungicide treatment (Figure S4).Pesticide treatments significantly enhanced survival, in a way that depended on time (Figure 1a; Tables S1-S8).We present the plot-level models (model set 1a, based on 3-6 seedlings/plot), though we note the individuallevel models (model set 2, based on individual seedlings) yielded identical results (Figure S5; Tables S9-S15).One week after the start of the experiment at the first observation point, there were no differences in survival between control and pesticide-treated seedlings ( 2 2 = 0.981, p = 0.612); 8 weeks after the experiment was established, survival of treated seedlings was 1.56 times higher on average than the control seedlings ( 2 2 = 12.48, p = 0.002; Figure 1b).This difference was statistically significant at a Bonferroni-corrected α = 0.007 for the last two sampling points (Tables S1-S7).However, there was no difference between the insecticide and insecticide+fungicide treatments (Figure 1).The Poisson model (model set 1b) confirmed the interaction between time and treatment, with the number of dead seedlings accumulating faster in control plots than treatment plots (Figure S6; Table S8).
There was no interaction between functional group and treatment (p > 0.05 for functional group × treatment interaction at all seven time periods), suggesting that the effects of release from enemies were consistent between C4 grasses and non-legume forbs.
However, there was a main effect of functional group, with survival of grass seedlings being 1.92 times higher than that of forbs 8 weeks after establishment ( 2 1 = 12.68, p < 0.001).We do not draw specific inferences from this pattern, as it could reflect different tolerances to transplantation or slight differences in time to germination between the functional groups rather than an ecological effect per se.

| Question 2: How do the benefits of release vary with community-level abiotic and biotic characteristics?
Community successional stage and species richness in plots were correlated with seedling survival.Based on the model at the final time point (Table S7), the proportion of seedlings surviving in plots at the early successional stage was 1.39 times higher than in mid successional plots ( 2 1 = 15.79,p < 0.001; Figure 1a; Figure S7a).Survival of seedlings also increased with plot species richness ( 2 1 = 9.95, p = 0.002; Figure 2; Figure S7b).These effects were consistent in magnitude and statistical significance throughout the growing season (Figure S7; Tables S1-S7).Combined with the increasing influence of treatment, our ability to explain variation in survival therefore increased throughout the growing season, with the conditional R 2 rising from 0.559 to 0.651 after 8 weeks.Seedling survival was not related to seedling-community phylogenetic distance, light availability or plot soil moisture at any time during the growing season (Tables S1-S7).

| Question 3: Can variation in the benefits of release be explained by species-level characteristics?
There was high variation in the effect of treatments among species and in overall seedling survival (Figure 3; Figure S8; Tables S16-S31).
We explored whether species-specific differences in effect sizes could be explained by phylogenetic relatedness among species, species traits or rates of damage by insects or fungi.
Effect sizes were not related to phylogeny, for either the insecticide (C mean = −0.125,p = 0.782) or insecticide+fungicide treatment (C mean = −0.217,p = 0.925) (Figure 4).Effect sizes were also unrelated to levels of damage on control plants (Figure S9; Table S32).However, aboveground leaf traits (Figure 5a) were correlated with effect sizes, with more resource-acquisitive plants showing higher benefit from the insecticide treatment in both functional groups (F 1,13 = 4.945, p = 0.045; Figure 5b; Table S33).This relationship was also present but non-significant for the insecticide+fungicide treatment.Neither

F I G U R E 3
The effect sizes of insecticide and insecticide+fungicide treatments relative to the control after 8 weeks (±95% CI), separated by functional group and species.The blue dashed horizontal line represents survival rates equal to the control group, averaged across all other factors.

F I G U R E 4
No statistically significant relationship between phylogeny of planted species and the effect size of insecticide treatment (left) and insecticide+fungicide treatment (right).Grasses are positioned above forbs.
leaf SLA nor leaf N were significantly correlated with effect sizes, but they showed the same trend, with high-SLA and high-leaf N species showing larger positive effect sizes in both the insecticide and insecticide+fungicide treatments (Table S33).There was no main effect of functional group in these analyses, supporting the results from Question 1 that the two functional groups have similar effect sizes in response to release from enemies.

| DISCUSS ION
By hand-painting individual seedlings with pesticides to enforce aboveground release from enemies, we found that enemy removal increased seedling survival of both non-legume forbs and C4 grasses in this field experiment by an average of 56%.Our result suggests enemies play an important role determining recruitment success into grassland communities.Our findings also support the key mechanism underlying the ERH.As we focused on 16 different species from two functional groups, we also demonstrate the wide scope of this result, emphasising the potential importance of early enemy regulation regardless of specific colonist identity or functional group.
4.1 | Question 1: Does release from enemies increase seedling survival in natural communities, and is this effect consistent between C4 grasses and forbs?
The beneficial effect of enemy release was broadly consistent across successional stages and functional groups (Figures 1 and 3).Effect sizes for insecticide and insecticide+fungicide treatments were the same (Figure 1): this may suggest that seedling survival is strongly reduced by insect herbivory but not fungal infection in this system.
Fungal pathogens may not reduce plant survival but instead limit biomass and competitive ability later in life (e.g.Mitchell, 2003;Zaret et al., 2022).Alternatively, it could suggest that fungal release provides no additional benefit if seedlings have already been released from insects.Based on our experimental design with no separate fungicide treatment we cannot rule out the possibility that this would also be true for fungi; in other words, it could be that release from insects would provide no additional benefit if they have already been released from fungi.
The survival benefit of release from insect herbivores was equally observed in both C4 grasses and non-legume forbs (Figure 3).The benefit to C4 grasses was somewhat unexpected, given previous work has found little benefit of pesticides on C4 grasses (Seabloom et al., 2018).However, this previous work focused on biomass rather than early survival, emphasising the importance of quantifying recruitment when considering enemy effects on communities.While in general fungal pathogens are more common on grasses and insect herbivores are more common on forbs (Ebeling et al., 2022;Fricke et al., 2022), a pattern also borne out in our experiment (Figure S9), C4 grasses may be particularly vulnerable as seedlings relative to adults (Dybzinski & Tilman, 2012) and thus also suffer reduced survival from insect tissue removal.

| Question 2: How do the benefits of release vary with community-level abiotic and biotic characteristics?
Seedling survival was influenced by plot successional stage and plotlevel community richness.Other studies have found that mid succession communities tend to be more species-rich and nutrient-poor than early succession communities, characteristics that increase resistance to invasion and colonisation (Catford et al., 2020;Fridley et al., 2007;Liu et al., 2021).Indeed, our mid plots had an average of (more resource-conservative) a species was, the lower the benefit from insecticide treatment (grey line and shading).For full PCA results including species labels, see Figure S1.
two more species per plot than early plots and total summed plant cover was 20% higher in mid plots, potentially exerting stronger competitive pressure on seedlings (Figure S7a).There was no interaction between community type and treatment (Figure 1a), suggesting the impacts of enemies on recruitment are reasonably consistent across community types.We only studied one early and one mid site (though with high plot-level replication), so can draw only limited inference about the interaction between enemies and successional stage.However, our plots are representative of local early and mid succession sites (Figure S3), and other recent work suggests that enemy effects and community effects (such as nutrient limitation) act independently rather than interactively to influence plant community assembly (e.g.Heckman, 2023;Shan et al., 2023;Zhao et al., 2023), supporting our conclusion.
Within each site we unexpectedly observed a slight increase in seedling survival as plot richness increased (Figure 2).This result has also been observed before, though in tree communities studied over longer time periods (Liu et al., 2022).We suggest that plotlevel richness may be correlated with reductions in total enemy pressure, as increased local diversity has been shown to decrease infection prevalences on individual plants (Rottstock et al., 2014).
While we observed no relationship between plot richness and aboveground damage rates in our study (Figure S4), it is possible that increased richness reduced the impact of unmeasured belowground pathogens.
Our results indicated that seedling survival was unrelated to light availability and soil moisture.Generally, increased light availability is expected to facilitate recruitment (e.g.Johnson et al., 2020).
However, nitrogen rather than light is the key limiting resource at Cedar Creek (Tilman, 1990).We thus expected that light availability would have a negligible effect in our study.We also observed relatively low variation in moisture across our plots, which may explain why moisture levels did not significantly affect seedling survival in our study.Further, recent evidence from Cedar Creek suggests that water availability is important largely in disturbed habitats, through its interaction with soil nitrogen (Catford et al., 2023).Therefore, while we found light and moisture to be relatively unimportant in our study, we suggest that the influence of these variables should be tested at sites with different conditions to Cedar Creek.

| Question 3: Can variation in the benefits of release be explained by species-level characteristics?
Aboveground traits partially explained variation among species: resource-acquisitive species showed stronger responses to release from enemies than resource-conservative species (Figure 5).
Resource-acquisitive species typically invest in fewer defences and are more vulnerable to enemies (Cappelli et al., 2020;Grutters et al., 2017;Heckman et al., 2019).Resource-acquisitive plants may therefore be especially damaged and should receive greater benefits from enemy release (Blumenthal et al., 2009).Resourceconservative plants should benefit little from enemy release as they are heavily defended so less susceptible to damage.However, on average, resource-conservative species still benefited in our experiment, possibly because juveniles of all species, even resourceconservative ones, are vulnerable to enemy damage to some extent (Bruns et al., 2022).Therefore, while the leaf economics spectrum highlights which species may be most regulated during recruitment, enemies may affect the establishment of all grassland species (Figure 5b).In contrast, our results suggest that phylogenetic predictors are not important in determining enemy regulation.There was no phylogenetically conserved response to release from enemies (Figure 4), nor did phylogenetic relatedness to the recipient community affect the response to enemies.The absence of a clear phylogenetic signal affirms the need to incorporate information on functional traits into predictions of recruitment success, as phylogeny alone may not always predict meaningful ecological differences (Cadotte et al., 2017).
Resource-acquisitive species benefited the most from release, but this does not seem to fit with our observation that effect sizes were unrelated to damage on control plants (Figure S8): we would expect that more highly damaged plants in the controls should benefit the most (e.g.see Cappelli et al., 2020).We suggest this lack of relationship is because we only assessed damage once, at the end of the season.Highly damaged plants may have already died, thus lowering average damage on (surviving) control seedlings and masking the expected relationship between damage and enemy removal benefits.

| Implications for the Enemy Release Hypothesis
Our results mechanistically link release from enemies with greater recruitment success, emphasising the possible benefits of enemy release shortly after the arrival of exotic plants.It has been argued that only vertebrate herbivores prevent establishment of exotic plants as they consume whole plants or parts of plants, and that nonvertebrate enemies only limit invader abundance or spread once populations are established (Levine et al., 2004;Scurr et al., 2008;Zhang et al., 2018).Our findings suggest that, by removing tissue, insect herbivores may have a similar impact on seedlings as vertebrate herbivores, as even small levels of tissue removal will have a disproportionately large impact on seedlings.Our results also support the hypothesis that resource-acquisitive species should benefit the most from release from enemies (Blumenthal et al., 2009;Brian & Catford, 2023), increasing their likelihood of establishing and successfully dominating invaded communities.Initial enemy release may therefore facilitate establishment of exotic species, even if enemies subsequently accumulate on exotic populations through time (Dong et al., 2018;Hawkes, 2007;Katz & Ibáñez, 2017; but see Xirocostas et al., 2023).
Few experiments of the ERH test survival of exotic plants or release early in the invasion process (Brian & Catford, 2023;Hawkes, 2007).In addition, experiments that use exotic species are often confounded both by different conditions between home and invaded ranges, and plot-level experimental treatments, making it extremely difficult to disentangle the effect of enemies from abiotic conditions (e.g.resource levels) and other biotic interactions (e.g.competition) (Agrawal & Maron, 2022;Brian & Catford, 2023;Catford et al., 2022;Pearson et al., 2022).In contrast, we examined the key mechanism of enemy release by applying pesticides to individual plants, and so were able to disentangle the effect of enemies on recruitment from those of competition or resources.While we used native seedlings and so do not directly test the full ERH, we suggest that the field of invasion biology would benefit from further experiments that apply pesticides to individual seedlings, to test whether the core mechanism of the ERH holds over a wider range of conditions than those explored here.Understanding how ecological context interacts with enemy release is vital to understanding the role of enemy release in global invasions (Brian & Catford, 2023).

| CON CLUS IONS
We have demonstrated that insect herbivores significantly affect recruitment into grassland communities, reducing the odds of seedling survival by an average of 56%.This result was consistent across eight C4 grass and eight non-legume forb species and for seedlings transplanted into communities at different successional stages, and was highest for resource-acquisitive species.This experiment provides strong evidence that invertebrate enemies play a significant role in community assembly but that the strength of this regulatory role depends on the traits of recruiting species.Further, by showing that release from enemies can significantly enhance the survival of seedlings, it suggests that enemy release-mediated recruitment may be an important but understudied aspect of plant species invasions.
Even if enemies rapidly accumulate on invaders (Hawkes, 2007;Ivison et al., 2023), brief release during plant recruitment may be sufficient for invaders to gain a foothold in their new ranges.
Xirocostas et al. (2022) to ensure accuracy in our estimates.All visual estimation for percent cover and damage was carried out by a single observer (J.I.B.).On 15 August 2022, we measured % soil moisture content in all 288 plots, taking the average of four measurements in each plot using a Delta-T Devices HH2 Moisture Meter and SM300 Soil Moisture Sensor.On 16 August 2022, we measured light availability in each plot by taking the average of three light readings below the canopy with a Decagon Sunfleck PAR Ceptometer, dividing this by the light level above the vegetation layer and multiplying by 100.

a
response variable, and running a generalised linear mixed effects model (Poisson family, log link) with a treatment × time interaction and successional stage, functional group, plot richness, phylogenetic distance, moisture and light availability as fixed effects, with species nested within functional group as a random effect (model set 1b, Methods S1).The fact that the number of seedlings available to die declines through time does not affect the detection of the treatment × time interaction (the target of this analysis) or the assumptions of a Poisson model.

F
Pesticides enhanced seedling survival.(a) The overall raw proportion of seedlings per plot that survived, separated by treatment and successional stage.Lines are smoothed loess fits ±95% CI (shaded).(b) Effect size of treatments at each time point, averaged across all species (±95% CI).The dashed blue horizontal line represents survival rates equal to the control group, averaged across all other factors.F I G U R E 2 The effects of plot richness (total number of species in the plot) on the survival of seedlings ±95% CI.This is a marginal effects plot from the final time point (8 August 2022), holding all other factors constant.Black line and shading indicate the overall richness effect that was consistent regardless of treatment; individual plots are coloured by treatment for reference and jittered to aid visualisation.

F
Relationship between leaf trait principal component analysis and effect sizes.Points show individual species.(a) The first two PC axes with associated traits.(b) The further along PC1

Figure S3 :
Figure S3: NMDS of taxonomic community composition of our early and mid-site plots and how they represent sites of that age at Cedar Creek.

Figure S4 :
Figure S4: Effectiveness of pesticide treatment for reducing insect damage (insecticide treatment) and for reducing insect and fungal damage (insecticide+fungicide treatment).

Figure S5 :
Figure S5: Effects of pesticide treatment on individual seedling survival relative to control seedlings.

Figure S6 :
Figure S6: The number of new dead seedlings per plot at each sampling event, split by treatment.

Figure S7 :
Figure S7: The effect of successional stage and species richness in determining the odds of survival of planted seedlings.

Figure S8 :
Figure S8: Proportion of seedlings of each species surviving per plot, separated by successional stage and averaging across all other factors.

Figure S9 :
Figure S9: Relationship between effect sizes for insecticide treatment, insecticide+fungicide treatment, and damage rates on control plants.Tables S1-S7: Results from model set 1a (quasibinomial models exploring proportional seedling survival at the plot level at each time point).

Table S32 :
Relationship between damage on control plants and treatment effect sizes.