Plant volatiles and priority effects interactively determined initial community assembly of arthropods on multiple willow species

Abstract Plant traits, which are often species specific, can serve as environmental filtering for community assembly on plants. At the same time, the species identity of the initially colonizing arthropods would vary between plant individuals, which would subsequently influence colonizing arthropods and community development in the later stages. However, it remains unclear whether interindividual divergence due to priority effects is equally important as plant trait‐specific environmental filtering in the initial stages. In this study, we propose that plant volatile organic compounds (PVOCs) may play a crucial role as an environmental filter in the initial stages of community assembly, which can prevent the community assembly process from being purely stochastic. To test this hypothesis, we conducted short term but highly frequent monitoring (19 observations over 9 days) of arthropod community assembly on intact individuals of six willow species in a common garden. PVOC compositions were analyzed before starting the experiment and compared with arthropod compositions occurring on Days 1–2 of the experiment (earliest colonizer community) and those occurring on Days 8–9 of the experiment (subsequent colonizer community). Unintentionally, deer herbivory also occurred at night of Day 2. Distance‐based statistics demonstrated that PVOC compositions were plant species specific, but neither the earliest colonizer nor the subsequent colonizer community composition could be explained by plant species identity. Rather, Procrustes analysis showed that both the PVOC composition and that of the earliest colonizer community can be used to explain the subsequent colonizer community. In addition, the linkage between PVOCs and the subsequent colonizer community was stronger on individuals with deer herbivory. These findings indicate that PVOCs have widespread effects on initial community assembly, as well as priority effects brought on by stochastic immigration, and that plant species identity only has weak and indirect effects on the actual composition of the community.

effects is equally important as plant trait-specific environmental filtering in the initial stages. In this study, we propose that plant volatile organic compounds (PVOCs) may play a crucial role as an environmental filter in the initial stages of community assembly, which can prevent the community assembly process from being purely stochastic. To test this hypothesis, we conducted short term but highly frequent monitoring (19 observations over 9 days) of arthropod community assembly on intact individuals of six willow species in a common garden. PVOC compositions were analyzed before starting the experiment and compared with arthropod compositions occurring on Days 1-2 of the experiment (earliest colonizer community) and those occurring on Days 8-9 of the experiment (subsequent colonizer community). Unintentionally, deer herbivory also occurred at night of Day 2. Distance-based statistics demonstrated that PVOC compositions were plant species specific, but neither the earliest colonizer nor the subsequent colonizer community composition could be explained by plant species identity. Rather, Procrustes analysis showed that both the PVOC composition and that of the earliest colonizer community can be used to explain the subsequent colonizer community. In addition, the linkage between PVOCs and the subsequent colonizer community was stronger on individuals with deer herbivory. These findings indicate that PVOCs have widespread effects on initial community assembly, as well as priority effects brought on by stochastic immigration, and that plant species identity only has weak and indirect effects on the actual composition of the community.

K E Y W O R D S
community assembly, plant-animal interaction, plant-arthropod feedback

| INTRODUC TI ON
Arthropod communities on plants are one of the model systems for community assembly research, with dynamics driven by interactions between arthropod populations and plant traits (acting as environments; Utsumi et al., 2010). Plant traits are often species specific (Hopkins et al., 2017;Lehrman et al., 2012;Moreira et al., 2014) and genotype specific (Barbour et al., 2015;Lehrman et al., 2013), which act as environmental filtering (Chen et al., 2017;Götzenberger et al., 2012;Zobel, 1997) for arthropod community assembly. This results in arthropod community composition depending on the plant species and genotype identity (Hochwender & Fritz, 2004). At the same time, induced responses such as induced resistance and compensatory regrowth are herbivore-species-specific and deterministic (Kessler & Halitschke, 2007;Van Zandt & Agrawal, 2004a), which in turn affect the subsequent community assembly (Stam et al., 2018;Van Zandt & Agrawal, 2004b), forming feedback between plant traits and arthropods (i.e., plant-arthropod feedback, Utsumi et al., 2010;Yoneya et al., 2023). When such feedback is initiated by the earliest colonizers on a plant whose identity can vary stochastically (Bacca et al., 2021;Shinohara & Yoshida, 2021), it results in a priority effect (Chase, 2003;Fukami, 2015) because of the combination of stochastic and deterministic processes (Chase, 2010), which contribute to the divergence of community composition even within a plant population.
Plant volatile organic compounds (PVOCs), a key plant trait, influence the behavior of arthropods, including both herbivores and their natural enemies, by acting as attractants and repellents (Yoneya & Miki, 2015 and references therein). PVOCs affect herbivore communities and their predator and parasitoid populations (Fatouros et al., 2012;Karban & Shiojiri, 2009;Morrell & Kessler, 2017;Xiao et al., 2012). Intriguingly, PVOC compositions are not only plant species-specific (Peacock et al., 2001;Yoneya et al., 2012), but also depend on the identity of herbivore species (Mann et al., 2021) and co-occurring plant species (Grof-Tisza et al., 2021;Yoneya et al., 2012) especially when induced by herbivory. For example, the composition of uninfested willow PVOCs differs significantly among cooccurring seven willow species, whereas the similarity of PVOCs induced by herbivory of the common leaf beetle species is high in some of these seven species and attracts its natural enemy predator more effectively than other willow species (Yoneya et al., 2012).
The current experimental approaches, which primarily focus on single or limited tri-trophic food chains within arthropod communities and rely on choice tests for individual species, are ineffective in understanding the roles of PVOCs in highly diverse arthropod communities. This limitation arises from the difficulty of conducting tests for every arthropod species, particularly minor ones. Moreover, the community composition cannot be explained by the superposition of each species' preference to PVOCs, as arthropod species interact with each other in various ways, including competition and preypredator interactions (Yoneya & Miki, 2015). Nevertheless, very few studies (Zu et al., 2020) have investigated the linkage between PVOCs and diverse arthropod communities. At the same time, the divergence of arthropod communities between plant individuals due to priority effects has also received increasing attention (Stam et al., 2018;Utsumi, 2015;Van Zandt & Agrawal, 2004b;Yoneya et al., 2023). Previous studies have addressed the role of initial stage communities in shaping later stages of community development by experimentally manipulating either the initial herbivore colonizers (Stam et al., 2018;Utsumi, 2015;Van Zandt & Agrawal, 2004b) or the plant initial conditions, including exposure to herbivore-induced PVOCs , while maintaining other environmental conditions homogeneous. The manipulation of initial colonizers in a homogeneous environment has been widely applied to microbial (Dickie et al., 2012;Toju et al., 2018) and plant (Fukami et al., 2005) communities. Consequently, the structuring of initial stage communities by dispersal and interspecific interactions in a heterogeneous environment remains unclear. Specifically, it remains underexplored how interindividual divergence resulting from priority effects interacts with plant trait-specific environmental filtering during the initial stages of community assembly, encompassing both stochastic and deterministic processes (Chase, 2010). In this study, we propose that PVOCs may play a crucial role as an environmental filter in the initial stages of community assembly, which can prevent the community assembly process from being purely stochastic. PVOCs can alter the behaviors of both herbivores and their natural enemies in a predictable way, affecting the mortality of herbivores as well as immigration and emigration of herbivores and carnivores, thereby acting as the environmental filter.
The overarching objective of this research is to contribute to a better understanding of the mechanisms underlying the initial stages of community assembly by investigating the proposed role of PVOCs as the environmental filter. We conducted short term (9 days from the establishment) but highly frequent (two-three times per day) monitoring of arthropod community assembly on four intact potted individuals (two clones × two individuals per clone) from each of the six willow species in a common garden. We sampled the headspace around the potted plant to analyze PVOC compositions from all individuals (24 in total) before starting the community assembly experiment and compared them with arthropod compositions occurring on Days 1-2 of the experiment (hereinafter, termed the earliest colonizer community) and those occurring on Days 8-9 of the experiment (termed the subsequent colonizer community). Deer herbivory occurred unintentionally during the second night of the community assembly experiment; therefore, we investigated its influence on the colonizer communities.

Community ecology
With this monitoring, we specifically aimed to test the overall connection between PVOCs, arthropod communities, and plant species identity, and to examine how the strength of the connection is determined. To achieve this, we combined distance-based (PERMANOVA and Mantel test) and raw-based (Procrustes analysis; Lisboa et al., 2014;Peres-Neto & Jackson, 2001) approaches.
Procrustes analysis, which is comparable to the Mantel test but offers greater statistical power for detecting the overall linkage between matrices (Peres-Neto & Jackson, 2001), not only allows us to assess the overall linkage but also provides a means to evaluate the strength of linkage for individual observations. Meanwhile, distance-based approaches have long been employed in community assembly studies (Anderson et al., 2011). By incorporating raw-based approaches alongside distance-based approaches, we can further untangle the complex relationships between PVOCs and arthropod communities. We developed three hypotheses that were not mutually exclusive to explore different factors influencing community assembly. First, we hypothesized that both plant species identity and PVOCs contribute to the community assembly. This hypothesis allowed us to examine whether PVOCs and other plant traits act as environmental filters. Plant species identity served as a proxy for unmeasured plant species-specific traits. Second, we hypothesized that the earliest colonizer composition also affects how the community is assembled later. This hypothesis aimed to investigate the significance of priority effects in the initial stages of community assembly, in addition to environmental filtering. Finally, we hypothesized that large herbivores can influence community assembly. Our monitoring design primarily focused on testing the first and second hypotheses, while the including of the third hypothesis aimed to provide a possible explanation for any unintended impacts of deer herbivory (Fraser et al., 2018). The abbreviations in the first brackets after scientific names are used in the following text and figures. The words in the second brackets are the corresponding Japanese names of the willow species, which would help avoid confusion related to the synonyms in willow taxonomy, since these are more stable and have fewer synonyms (Ohashi, 2000). The natural populations of these six species, as well

2002.
We cut and collected 1-2-year-old branches of willows (two branches per clone × two clones × six species) from the willow garden in the CER on May 13, 2016. After keeping their basal section in water for 10 days until root emergence, we potted them individually in soil (Takii potting soil, Takii & Co., Ltd.) in pots (diameter 9 cm, height 7.6 cm) and supplied them with fertilizer (Hyponex (N-P-K = 6-10-5); HYPONeX) every 2 weeks. Potted plants were maintained in a greenhouse under a long-day photoperiod (18 L: 6 D) at 25 ± 2°C and 50%-70% RH until the newly emerged shoots were approximately 15-20 cm in height with approximately 20 leaves. We used four potted plants for each willow species (two clones × two replicates per clone). These potted plants were used to collect plant volatiles and then located randomly in three lines of eight pots at 2-m interval in the experimental garden next to the willow garden in CER on June 21, 2016. It appears that replicates within each clone are pseudo-replicates. However, the high vegetative reproduction ability of floodplain species of willows (Radtke et al., 2012) can result in the dominance of a single clone in multiple populations along a single river (Beismann et al., 1997), leading to neighboring willow individuals in a short distance being genetically identical in natural environments. Therefore, including multiple individuals from a single clone as independent replicates was reasonable to investigate the roles of stochastic colonization and priority effects in an identical environment (identical genetical background), as well as environmental filtering (from different plant clones and species) in community assembly processes. Differences resulting from plant clones were not our main focus but treated as a nested factor within plant species in the statistical analyses.

| PVOC characterization
The volatiles were collected from each potted plant in a climatecontrolled room. A hexane solution of tridecane (0.5 mg/mL) impregnated into a piece of filter paper (1 cm 2 ) was used as the internal standard. The potted willow plant was covered with a TPS bag (500 mm × 350 mm; Mitsubishi Gas Chemical Co.), and closed at the base of its stem with a clip. The TPS bag was fitted using two nozzles. One nozzle was connected to an air cylinder and the other was connected to a glass tube packed with an adsorbent (Tenax TA 60/80, Gerstel GmbH & Co. KG). The glass was connected to an air pump (Sibata Scientific Technology, Ltd.). Purified air from the cylinder was directed into the TPS bag, and volatile compounds from the headspace of the plant were collected in the adsorbent by generating an air flow (100 mL/min) using an air pump collected on the adsorbent for 40 min.
The collected volatile compounds were analyzed using a gas chromatograph mass spectrometer (6890/5972 GC-MS (mass selective detector, 70 eV), Agilent Technologies, Santa Clara, CA, USA) equipped with an HP-5MS capillary column (Agilent; length 30 m, ID 0.25 mm, film thickness 0.25 mm). The system was equipped with a thermal desorption system (TDS), cooled injection system, and cold trap system (CTS, GERSTEL GmbH & Co. KG). Headspace volatiles collected on the adsorbent were released by heating the TDS at 280°C for 4 min. Flash heating of the cold trap unit induced sharp injection of the compounds into the capillary column of the GC. The GC oven temperature was programmed to increase from 40°C (9-min hold) to 280°C at a rate of 5°C/min. The headspace volatiles were tentatively identified by comparing their mass spectra with those from Wiley databases (Wiley7N and Wiley275) and the database of the National Institute of Advanced Industrial Science and Technology (SDBS compounds and spectral search; http://riodb 01.ibase.aist.go.jp/sdbs/cgi-bin/direct_frame_top.cgi). The retention times of the volatiles were further compared with those of standard compounds (Wako Chemicals, and IFF Chemical). Compounds for which no standards were available were regarded as tentatively identified when >90% of spectra matched those in the databases.

| Community monitoring
The number of arthropods found on each plant was surveyed from Day 1 when the potted willow plants were placed in a common garden and was observed for 9 days (at 9 AM, 12 PM, and 3 PM on Day 1-2, and at 9 AM and 3 PM from Day 3-8 and at 9 AM at Day 9 (June 29, 2016)).

| Deer herbivory
The shoots of potted willow plants were found to be consumed by deer (Cervus nippon) early in the morning before the survey on June 23 (Day 3 of the experiment). The potted willow damaged with deer herbivory was recorded (11 pots in total: Table A1). After that, a plastic net with a 10-cm mesh was covered around a plot where potted willow plants were placed (1 m away from the outermost pots). A trail camera was installed around the experimental plot in the common garden, and photos captured deer walking around the willow garden early morning on June 24, indicating that the damage on the night of Day 2 was likely due to deer herbivory.

| Data preprocessing
All statistical analyses and visualizations were conducted using R4.2.0 (R Core Team, 2022). The raw datasets of community monitoring for 9 days with multiple measurements within a day were first separated into each day subsets. On this short-term scale, it is reasonable to assume that mortality, immigration, and emigration may play a larger role in arthropod population dynamics than local birth processes on each plant. In addition, the appearance of an arthropod individual at each measurement time point could be just a transient stay on the plant. For each of the colonizer communities from Day 1 to Day 9, the cumulative number of individuals observed for each species was calculated to reflect the persistence of each individual stay and regarded them as community compositions at each day. We calculated the 2-day moving sum of the abundance of each observed species (i.e., combining data from consecutive days, such as Days 1 and 2, Days 2 and 3, up to Days 8 and 9, for the colonizer communities), in order to reduce sparseness in the community compositions.
These two types of communities (communities of each day and those at 2-day scale) were used for further statistical analysis. For the analysis of temporal dynamics of populations and communities, we primarily utilized population size and community composition data of each day. To compare the differences between the beginning and end of the 9-day time series, we focused on the earliest colonizer community (Days 1-2 community) and the subsequent colonizer community (Days 8-9 community).

| Basic analysis of populations and communities
Prior to applying multivariate statistics to the community compositions, we summarized the temporal dynamics of key parameters in the communities, including the total abundance of colonizers, taxonomic richness, and population sizes of major taxa. We generated the rank abundance curve by cumulatively adding up the abundance of each taxon and then selected the top five taxa to examine their population dynamics. We used the gls function in the nlme package, which incorporates an autoregressive structure of order one (i.e., AR(1)), for a repeated measures ANOVA to examine the relationship between explanatory variables (plant species identity, deer herbivory, PVOCs, and time; See also Section 2.5.5) and response variables (the total abundance of colonizers, taxonomic richness, and population sizes of major taxa).

| Distance-based approaches
The zero-adjusted Bray-Curtis dissimilarity for community composition, which is more appropriate than Bray-Curtis dissimilarity for sparse samples in the initial phases of community establishment (Clarke et al., 2006), and the Hellinger dissimilarity for PVOC composition were used to convert the raw composition data into distance matrices. To confirm whether the presence/absence patterns of each taxon differed between samples, we first binarized the composition matrix and then applied the zero-adjusted Bray-Curtis dissimilarity, since Bray-Curtis dissimilarities are influenced by differences in abundances even when the presence/absence patterns are identical.
When examining the temporal changes in community composition on each plant individual, we calculated the dissimilarities between the community compositions at each day using two methods.
The first method involved calculating the dissimilarities between the community composition on Day 1 and on each day thereafter (from Day 2 to Day 9). The second method involved comparing the communities at each day with those on the previous day to calculate the dissimilarities between neighboring day communities. Since raw temporal series of compositional data were converted into temporal differences, we did not consider AR(1) structure and simply applied linear model to these datasets when investigating the influences of plant species identity and deer herbivory. More specifically, we first prepared the full model (divergence ~ days + plant/plant clone + deer_arrival + days: plant + days:deer_arrival + plant:deer_arrival) and then selected the best model by stepAIC function.
When testing whether each of these compositions (i.e., PVOC composition, earliest colonizer community, and subsequent colonizer community) differed between categorical parameters (e.g., plant species identity and the presence/absence of deer herbivory), PERMANOVA (vegan::adonis) was employed after checking for the absence of heterogeneity of divergence (dispersion) within a level through PERMDISP (vegan::betadisper), with 4999 permutations. This is the same when checking if the community composition differed between days only when homogeneous dispersions between days were confirmed. Note that we also investigate the relationship between deer herbivory (occurring at the night of day 2), PVOC composition (evaluated before the experiment), and earliest colonizer community (observed in day 1-2) in order to confirm whether deer herbivory occurred randomly or was related to plant traits or conditions. The Mantel test (vegan::mantel) was also used to test for the presence of an overall linkage between two distance matrices.
Plant volatile organic compound composition in our study represents one of the constitutive plant traits since PVOCs were collected from plants without any herbivory. In order to quantify the degree of the inducibility of PVOCs, we also used the PVOC composition data from our previous study (Yoneya et al., 2012). We collected PVOCs from four replicates of each species intact individuals without herbivory and those infested with leaf beetle (Plagiodera versicolora). The dissimilarity between PVOCs was calculated by the Hellinger distance as the same as our main analyses and the spatial median of replicates was obtained by using betadisper() function in vegan package of R. Distance between the spatial median of the intact individuals and that of the infested individuals represents the inducibiliy of PVOCs for each willow species. Note that the source plant individuals for this PVOCs analysis were not identical to those in our study but originated from the same population (see Yoneya et al. (2012) for more details of PVOC collection).

| Raw-based approaches
We relied on Procrustes analysis, following the standard procedure for ecological datasets (Lisboa et al., 2014;Peres-Neto & Jackson, 2001). We directly compared the linkage between two raw matrices (vegan::procrustes) using the first two or five axes of principal coordinate analysis (PCoA) and checked its statistical significance with 4999 permutations (vegan::protest). This Procrustes test is comparable with Mantel test but has greater power for detecting the linkage between matrices (Peres-Neto & Jackson, 2001). We also used the Procrustes residual value (univariate) as an indicator of the deviation (or Procrustean association metric, Lisboa et al., 2014) of each observation (each plant individual in this study) from the overall linkage between two matrices, for further analysis of variance (ANOVA). To investigate the relationship between the first and final 2-day communities (i.e., the earliest colonizer community vs. the subsequent colonizer community), we also used partial Procrustes analysis to remove the effects of PVOCs on these communities. We constructed three matrices from the PCoA scores of PVOCs, the earliest colonizer communities, and the subsequent colonizer communities, and then applied the partial Procrustes analysis to these matrices. Since the existing function for partial Procrustes analysis (SINCSA::procrustes. partial) only returns the partial Procrustes correlation but does not the Procrustes residual values, we created our own custom function. This first uses res_cal function in gustave package to calculate the residuals of the community matrices after regressing them against the PCoA matrix and then uses these residuals as the input of procrustes function in vegan package.

| Overview of statistical variables
The statistical analyses in our study incorporated various response and explanatory variables, each serving a specific purpose. In the variables. Both Mantel test and Procrustes test utilized multivariate dissimilarity matrices from PVOC composition and arthropod community composition as both response and explanatory variables. In ANOVA for Procrustes residuals, the response variables were continuous univariates (Procrustes residuals), while the explanatory variables were categorical (plant species identity, plant clones, and deer herbivory). Notably, as this ANOVA was unrelated to time, the explanatory variables did not include time. In all analyses except for PERMDISP, plant clones were treated as a nested factor within plant species identity.

| Data visualization
Both the base plot functions and ggplot2 packages were used.

| Observed diversity of PVOCs and arthropods
Twenty-eight VOCs in total were identified from the headspace Moreover, 19 arthropod species/group (12 herbivores, seven predators/parasitoids) and one snail species (noting that a single individual was observed only once, although it was not an arthropod but was included in the community composition) were identified ( Table 1).

| Basic analysis of populations and communities
Based on the rank abundance curve ( Figure A1), we identified top five abundance taxa: leaf_beetle1 (Plagiodera versicolora), thrips (thrips sp.), aphid1 (Chaitophorus saliniger), aphid2 (Aphis farinosa), and spider (spider). Differences in their population dynamics between plant individuals were explained by plant species identity for two aphid species (C. saliniger and A. farinosa), by plant clones for two aphid species and spider, and by deer herbivory for A. farinosa ( Figure A2; Table A2).
Differences in the total abundance of colonizers in the commu- In addition, taxonomic richness was greater on plant individuals with deer herbivory at the 1st day (post hoc linear model p = .023) but became smaller than plant individuals without herbivory after deer arrival at the 2nd day night (Figure 1d).

| Distance-based approach
Variations in PVOC composition between individual plants were explained by plant species identity (nested PERMANOVA, df = 5, F = 4.005, p = .001) but could not be evaluated for plant clones due to heterogeneous dispersion (PERMDISP, df = 11, F = 9.458 × 10 29 , p = .001; Figure 2). The first two axes of PCoA of PVOC composition were also used in the linear model for the basic analysis of populations and communities but there were no statistically significant associations with five abundant taxa, total abundance, and taxonomic richness (p > .05).
The community development patterns in 1-and 2-day scales were not visually captured by simple PCoA plots (Figures A3 and A4).   Figure 3a). In addition, the divergence was lower when deer herbivory was present (df = 1, F = 5. 4493, p = .020724; Figure 3b). The similar patterns were obtained when evaluating the divergence between 1st day community and each day community ( Figure A5; Table A3).
The difference between the centroid positions of earliest colonizer community and subsequent colonizer community was not clear Similarly, the subsequent colonizer community ( Figure A6b There was also no influence of deer herbivory (PERMANOVA, df = 1, F = 1. 3536, p = .196). The overall correlation evaluated using the Mantel test was significant (r = .178, p = .009) for the earliest colonizer F I G U R E 1 Temporal development of total abundance and taxonomic richness. Total abundance (a, b) and the taxonomic richness (c, d) depending on plant species and the presence/absence of deer herbivory, respectively.
However, the subsequent colonizer community was explained by  Figure 6c).

F I G U R E 2 PVOC composition depending on plant species identity and plant clones. (a)
Hierarchical cluster tree to clarify the clustering of two plant individuals from the identical intraspecific clones. (b) PCoA plots to clarify the entire compositional differences. The first two axes of PcoA plot explained 23.8% and 13.79% variations, respectively.

| Impact of PVOCs on community assembly
PVOCs were an effective predictor of overall arthropod community assembly, as suggested by Procrustes analysis. Although the Mantel test did not detect the linkage between PVOCs and communities, we assume that Procrusted analysis have a greater statistical power (Peres-Neto & Jackson, 2001). Despite the pattern that PVOCs were plant species-specific, plant species identity itself explained a more limited range of community characteristics, such as population dynamics of a few abundant taxa, taxonomic richness, and the temporal development of the community. The species specificity of PVOCs described in this study (Figure 2) is consistent with recent studies on willows (Yoneya et al., 2012) as well as vegetable species (Yoneya et al., 2018). Similar to the genotype specificity of other plant traits (e.g., phenolics) of willow (Lehrman et al., 2012), willow PVOCs were also genotype-specific. Thus, the PVOCs of uninfested plants can be interpreted as a major constitutive plant trait and act as an environmental filter for community assembly. However, the commu-  Figure A6b).
Our results indicated that the environmental filtering function of PVOCs did not immediately become apparent when arthropods were allowed to colonize intact plant individuals. The effects of PVOCs as attractants for arthropod immigrations are weak and complicated by the stochastic behavior of organisms in very short term (within 2 days). However, PVOCs would influence the colonizer's choice of whether to stay on the plant individual, and the gradual increase in taxonomic richness and abundance on an individual plant made the link between PVOCs and community composition more apparent (within 9 days). This could explain why the link between PVOCs and community assembly was statistically significant for the subsequent colonizer community, but not for the earliest colonizer community.
Although not statistically significant, the linkage between PVOCs and the subsequent colonizer community was highly variable among plant species (Figure 6a). PVOCs in our experiment can be interpreted as one of the constitutive traits of plants because they were collected and quantified from intact (uninfested) plant individuals. One would hypothesize that plant species with greater inducibility to herbivory shift PVOC composition after herbivory to a greater extent, which in turn affects the subsequent assembly, weakens the linkage between the constitutive PVOCs and community composition, and leads to a greater inter-individual divergence of community composition. In fact, when re-analyzing the data from a previous study (Yoneya et al., 2012), there were interspecific variations in the degree of dissimilarity between constitutive PVOCs and leaf beetle-induced PVOCs (Figure 7a). However, there were no linkages with Procrustes errors (Figure 7b, linear regression, p = .178, adjusted R 2 = 0.039) and with interindividual divergence (Figure 7c, linear regression, p = .375, adjusted R 2 < 0). Although accumulating evidence suggests that induced PVOCs act as attractants and repellants for predatory and parasitoid arthropods as well as herbivorous arthropods (Beyaert & Hilker, 2014;Yoneya et al., 2010Yoneya et al., , 2012; it remains unclear whether induced PVOCs are important determinants of arthropod community assembly.

| Roles of unmeasured plant traits in community assembly
There is accumulating evidence suggesting the importance of chemical and physical defense in structuring arthropod communities. Specifically, the composition of secondary metabolites and physical defense levels are species-specific constitutive traits in Salicaceae species, which are linked to herbivore composition (Volf et al., 2015). However, such unmeasured plant species-specific constitutive traits were unlikely to influence the earliest and subsequent colonizer communities, since these communities were not clustered by plant species identity in this study. Similarly, the lack of linkage between the subsequent colonizer community and plant species identity implies that plant species-specific induced defense was unlikely to influence the community assembly. Induced defense also depends on the types of herbivory (chewer vs. sucker), and is F I G U R E 5 Procrustes residuals distribution. The ordination of the response variables (Y matrix) was rotated to minimize the squared difference with the ordination of the explanatory variables (X matrix). The matrix with the boxed texts is the rotated Y matrix. This represents the component of Y that is explained by X, and the endpoint of the blue arrows represents the corresponding point of the X matrix. (a) X and Y are PVOCs and the earliest colonizer community, respectively. (b) X and Y are PVOCs and the subsequent colonizer community, respectively. (c) X and Y are the earliest colonizer community and the subsequent colonizer community, respectively. (d) Linear relationships between communities and PVOCs was first partial out and then X and Y are set as the residuals of the linear regression with PVOCs of the earliest colonizer community and those of the subsequent colonizer community, respectively.
controlling PVOCs related to priority effects (Stam et al., 2018). P. versicolora, common herbivore in Salicaceae species, initiates the jasmonic acid pathway (Liu et al., 2022) and would negatively affect the late emerging conspecifics' performance (Yoneya et al., 2014) and its population dynamics . Additionally, at the interannual scale, willow species in floodplain often show compensatory regrowth after herbivory and physical damages, leading to greater resource availability and better nutritious quality (e.g., lower C:N ratio) for herbivores (Nakamura et al., 2005(Nakamura et al., , 2006. These herbivory-specific induced traits may have depended on the composition of the earliest colonizer community. Therefore, the presence of a linkage between the earliest and subsequent colonizer communities, even after controlling for the influences of PVOCs, suggests a potential role for these induced traits in shaping the subsequent colonizer community, despite not being measured in our study. In contrast, the earliest colonizer community was more likely to be structured stochastically, or theoretically influenced by unmeasured traits that are not plant species-specific nor PVOC-correlated.

| Priority effect
Our results suggest that both the priority effect and environmental filtering play significant roles in the observed community assembly processes. Linkage between the earliest and subsequent colonizer communities suggests the presence of priority effects caused by plant-arthropod feedback (Utsumi et al., 2010;Yoneya et al., 2023).
One may suspect that the high linkage from the Procrustes analysis was obtained simply because the community composition did not change in such a short period. However, this is unlikely because the presence/absence patterns of species changed on all plant individuals ( Figure 4c). Another possible explanation for the linkage between the earliest and subsequent colonizer communities is that they shared a common influence of PVOCs. Even after controlling for the influence of PVOCs, a significant correlation between the earliest and subsequent colonizer communities remained, suggesting the presence of priority effects. However, the specific mechanisms behind this effect, including whether indirect interspecific interactions  A priority effect is stronger with a stronger "herbivore specificity" of the herbivore-induced plant traits, that is, when the dependence of herbivore-induced plant traits on herbivore species identity is stronger. One may hypothesize that the inducibility of PVOCs shown in Figure 7a can be used as a loose proxy, but it is not.
While the inducibility of PVOCs represents the differences between constitutive PVOCs and those induced by a single herbivore species (Plagiodera versicolora), herbivore specificity should be evaluated using the differences in the induced traits between those induced by different herbivore species. The linkage strength between the earliest and subsequent colonizer communities was not associated with the inducibility of PVOCs (linear regression, p = 0.5383, adjusted R 2 < 0). impacts on resource availability, but such suppression occurred without negative impacts on the total abundance of arthropods ( Figure 1a) and four major taxon groups (except for aphid2; Figure A2). This suggests that the decline of architectural complexity, such as branching structure complexity, might hinder the colonization of new species (Marquis et al., 2002;Rudgers & Whitney, 2006), once the community composition was determined by the preference to PVOCs.

| Impact of deer herbivory
Mammals (e.g., elephant and wallaby) as well as arthropods use PVOCs as foraging cues (McArthur et al., 2019;Stutz et al., 2016). In light of this, the associations between deer herbivory, PVOCs, and arthropod communities may also imply the effects of PVOCs and arthropod communities on deer behavior. However, PVOCs measured before the assembly experiment, which represent constitutive plant traits, were not grouped by the presence/absence of deer herbivory (PERMANOVA,df = 1,F = 1.6856,p = .0686). Since the earliest colonizer communities were not grouped by the presence/absence of deer herbivory, it is unlikely that plant traits were modified by the earliest colonizers and then attracted deer. Notably, we observed a higher taxonomic richness on plants at the 1st day that would later be subjected to deer herbivory at night on the 2nd day ( Figure 1). This observation, however, remains inconclusive because our experiment was not intended to test the impact of plant conditions on deer behavior. Therefore, we assumed that the arrival of deer occurred as a stochastic event, while the community assembly after deer arrival (during the night of Day 2 of the experiment) was affected by deer herbivory.
Long-term (months to years) studies suggest that deer herbivory can induce changes in plant physical and chemical traits (Bryant, 2003;Den Herder et al., 2004;Stephan et al., 2017;Tanaka & Nakamura, 2015), but it remains unclear what can happen in the short term (9 days). To the best of our knowledge, there have been no studies on mammal-induced PVOCs. However, it is likely that deer herbivory induced changes in PVOCs during our experiment, because the induction of PVOCs after herbivory or mechanical damage typically occurs on a very short timescale (starting within minutes and hours and continuing for several days; Loreto et al., 2006;Martin et al., 2003;Staudt & Lhoutellier, 2007).
Deer herbivory destroys a greater part of plant tissues at once than arthropod herbivory, and its impact on plants would be similar to mechanical damage, which is known to increase the quantity as compensatory regrowth, alter the composition of PVOCs, and maintain species specificity (Peacock et al., 2001). We hypothesized that deer herbivory increased PVOC emissions while maintaining species specificity, allowing PVOCs to more effectively serve as behavioral cues for arthropods.
This also strengthened the relationship between PVOCs and the subsequent colonizer community by enhancing the environmental filtering roles of constitutive PVOCs in community assembly.

| Concluding remarks
In conclusion, our study found that both PVOCs and priority effects played a significant role in the short-term, initial stages of community assembly of arthropods on willow trees. Although our monitoring was limited to the initial stages, we predict that the effects of community divergence observed in our study was not transient but could persist in a longer timescale, at least for several months, based on previous studies (Stam et al., 2018;Yoneya et al., 2023) indicating such lasting effects. Our study indicated that riparian willow forests are an excellent model system for community assembly, along with recent studies focusing on the local co-occurrence of multiple willow species (Yoneya et al., 2012), its temporally dynamic characteristics (Nakamura et al., 2005), plant volatile-mediated food chains (Yoneya & Takabayashi, 2013), and high diversity of arthropods (Nakamura et al., 2006;Yoneya et al., 2023). We would suggest three directions for future research, in addition to longer monitoring of community assembly. First, larger spatial-scale experiments will elucidate the roles of spatial structure (or dispersal limit; e.g., variation partitioning and spatial clustering, (Coccia & Fariña, 2019)). The distribution of herbivore populations is substantially affected by the diffusibility of PVOCs and the induced responses of locally neighboring plant individuals around damaged trees (i.e., plant-plant communication; Ida et al., 2018;Morrell & Kessler, 2017), which will further influence community assembly. Second, community phylogenetic structure analysis will contribute to evaluate the relative importance of stochastic and deterministic forces in community assembly (Granville et al., 2023;Stegen et al., 2013). Third, herbivore specificities of the induced PVOCs, other defense chemicals, and plant regrowth are key indicators of priority effects via herbivore-specific-induced plant traits. These three directions will better predict the relative importance of dispersal limit, environmental filtering, and priority effects in shaping community composition and species diversity on plants.

CO N FLI C T O F I NTER E S T S TATEM ENT
The authors declare no competing interests.

DATA AVA I L A B I L I T Y S TAT E M E N T
The html generated from the R notebook for basic statistics, all statistical results, and graphics ("VOC_initial_migrator2022_shared.nb.html") as well as the datasets of PVOCs ("pvoc_en.csv". "7spp_volatile_rawdata.csv") and community composition ("grouped_raw_d_en.csv"), which were used in the R notebook, and the list of taxonomy information of arthropods ("arthropod_list.xlsx") are available in the figshare,

SPATIAL DISTRIBUTION OF WILLOW POTS
See Tables A1-A3 and Figures A1-A6 TA B L E A 1 Spatial distribution of the willow pots used for our experiment.