Biotic interactions shape trait assembly of marine communities across time and latitude

Abstract

Assembly processes are highly dynamic with biotic filters operating more intensely at local scales, yet the strength of biotic interactions can vary across time and latitude. Predation, for example, can be stronger at lower latitudes, while competition can intensify at later stages of assembly due to resource limitation. Since biotic filters act upon functional traits of organisms, we explored trait-mediated community assembly in diverse marine assemblages from four regions along the Pacific coast of North and Central America. Using predator exclusion experiments and two assembly stages, we tested the hypotheses that non-random trait patterns would emerge during late assembly at all regions due to competition and at lower latitude regions regardless of assembly stage due to predation. As expected, trait divergence occurred in late assembly but only at higher latitude regions, while in tropical Panama, relaxed predation caused trait divergence during late assembly. Moreover, colonizing trait strategies were common during early assembly while competitive strategies were favoured during late assembly at higher latitude regions. Predation-resistant traits were only favoured in Panama during both assembly stages. Our large-scale manipulative study demonstrates that different biotic interactions across time and latitude can have important consequences for trait assembly.

1. Introduction

Processes that structure communities act upon trait characteristics or functional traits of individuals, and at local scales the trait assembly of communities is primarily shaped by species interactions [1]. The strength of biotic processes, however, can vary geographically and have disproportionate effects in regions where they are expected to be stronger [2]. Predation can have greater influence on community structure, including patterns of biomass, composition and taxonomic and functional diversity in tropical areas [3–5], while competition can shape assembly at either high or low latitudes [6,7]. Many trait-based assembly studies consider competition as the only relevant biotic filter (sensu [8]), and few have investigated the influence of multiple biotic interactions in the assembly of communities across latitude (e.g. [9]). Therefore, despite strong evidence that both competition and predation are important structuring forces, our understanding of how these fundamental processes simultaneously shape trait-based community assembly across regions from different latitudes is largely unexplored (but see [10]).

In addition to geographical variation, the influence of biotic interactions may also depend on assembly stage. In sessile communities, for example, initial assembly may be random, and deterministic processes operate only when resources such as space become scarce [11]. Depending on the rate of community development, space saturation and the onset of competition can shift with time [12,13], and the strength of competition may increase later in development [12,14]. The influence of competition in structuring communities could become relevant at later stages of community development, highlighting the importance of exploring temporal shifts in community assembly.

To gain insight into the mechanisms that drive community assembly, the distribution and structure of functional traits can be used to elucidate underlying processes. A non-random distribution of traits within local assemblages suggests biotic processes are operating [15], however, abiotic heterogeneity can impact coexistence and obscure biotic processes when present at local scales [16]. With biotic interactions, greater trait dissimilarity than expected from random, or trait divergence, is predicted under strong resource competition as niche differentiation allows a greater number of species to coexist [8]. Alternatively, greater trait similarity, or trait convergence, can occur when shared traits optimize survival under strong predation pressure [17], or when competitive exclusion ensues [18]. Prior research examining trait distributions to infer community assembly processes has primarily been done in plant assemblages [8,19] or other monophyletic communities such as fish [9,20,21], zooplankton [17], or phytoplankton [22]. Further study is needed to determine whether biotic filters also shape the trait distribution of communities with greater functional and phylogenetic diversity. In this study, we explore trait-mediated assembly using marine invertebrate communities in which up to seven phyla can be represented across small spatial scales (tens of centimetres). We, therefore, test the generality of biotic interactions shaping trait distributions in an exceptionally diverse system.

The functional structure or mean traits of communities can provide further support for the influence of biotic processes beyond trait distribution alone and can identify optimal or unsuitable traits in a given scenario [5,23]. Plant traits related to drought resistance, for example, tend to decrease in wetter areas as competition for faster growth prevails [23]. Moreover, traits related to susceptibility to consumers (e.g. nutrient content, seed mass, body size, etc.) change in response to consumer pressure [5,24]. Changes in mean traits in addition to non-random trait distributions can help disentangle the influence of different biotic interactions at distinct latitudes and stages of community development.

We assessed the trait diversity and mean traits of sessile marine invertebrate communities developed under ambient or reduced predation over two assembly time stages (i.e. early or late) to evaluate the influence of biotic interactions as mechanisms of assembly across four regions and 47° latitude. We tested the hypotheses that trait convergence would emerge at regions located closer to the tropics regardless of assembly stage due to stronger predation. Moreover, trait divergence could become apparent during late assembly at all regions due to stronger competition. Using standardized field experiments in four regions, from the subarctic to the tropics, along the Pacific coast of North and Central America, we tested for patterns of trait convergence and divergence and changes in functional structure (i.e. optimal trait strategies) that underlie these patterns. Our study uses a novel combination of large-scale manipulative field experiments and trait-based modelling to provide new insights into how both predation and assembly time, as a proxy for competition, shape natural communities across a large geographical gradient. To our knowledge, this study is the first of this scale to investigate changes in trait responses to biotic interactions in functionally and phylogenetically diverse communities.

2. Methods

(a) . Study system

Sessile marine invertebrate communities are a tractable model system to examine fundamental assembly processes at local scales and across geographical gradients [25]. These near-shore communities are distributed across continents which facilitate regional comparisons [3,26]. Across a wide range of latitudes and within weeks, hundreds to thousands of pelagic invertebrate larvae can recruit on small substrate areas (i.e. 100 cm²) in near-shore habitats [27], and their fast growth enable the evaluation of assembly at multiple community development stages within a single year [12,25]. These communities are also functionally and phylogenetically diverse, with up to seven phyla (e.g. Chordata, Bryozoa, Annelida, Porifera, Mollusca, Arthropoda and Cnidaria) represented at the scale of centimetres [27]. Their wide array of traits related to life-history, defence strategies and competitive abilities can inform our understanding of trait-mediated assembly [5]. We thus used the distribution of functional traits to study the influence of biotic interactions on the assembly of sessile communities from four regions along the Pacific coast of North and Central America (electronic supplementary material, appendix S1, as published in [27, appendix S1]): Ketchikan, Alaska, USA (55°N, 131°W); San Francisco, California, USA (37°N, 122°W); La Paz, Baja California, Mexico (24°N, 110°W); and Panama City, Panama (8° N, 79°W).

To test for the effect of predators and time on the distribution of traits, we conducted a predator exclusion experiment across latitude for up to 1 year. In each region, we deployed experiments during periods of high productivity to capture peak growth and/or recruitment, corresponding to upwelling months in the tropics (December to April) and summer months at higher latitudes (electronic supplementary material, appendix S2). Invertebrate communities developed for three or 12 months to capture early and later assembly stages, respectively. In these communities, space is usually limiting by 12 months, therefore, we quantified the availability of open space as a limiting resource in each community. To standardize for habitat type, area, and assembly history, we used polyvinyl chloride (PVC) settlement panels (14 × 14 × 0.95 cm) as experimental substrate for invertebrate communities to assemble. Settlement panels provide a widely used model habitat in near-shore communities that facilitates both experimentation and in situ natural settlement, growth and interactions, similar to those observed on natural substrates [4,28].

We employed three treatments to modify predation pressure on the experimental communities: (1) cage exclosure; mesh size 0.635 cm² to exclude macropredators (i.e. fish) and reduce predation, (2) open; ambient predation and (3) partial cage; open panel surrounded by four sides of 0.635 cm² mesh as a procedural control that retained full access to predation [27]. We hung experimental panels 1 m below the water surface from floating docks at three recreational marinas in each region (i.e. 12 sites total) (electronic supplementary material, appendix S1, as published in [27, appendix S1]), and invertebrate communities developed facing the seafloor to minimize algal growth. In each region, assistants were hired and trained to perform cage cleaning and maintenance of experimental treatments following standardized protocols. Every two weeks, cages were scrubbed from the outside and inside when necessary to remove settled organisms, and cages with structural issues were replaced on site. Temperature and salinity measurements were recorded at 1 m depth from four opposite corners at each site every two weeks. In Panama, experimental panels for the 12-month experiment from one site were re-deployed at a later date due to losses to wave action and storms. Additionally, caged and partial caged treatments were not deployed for the late-stage assembly interval in Mexico due to logistical constraints. We retrieved five panels per treatment (i.e. cage, open, partial cage) per assembly time (i.e. early, late) at each of the three sites in the four regions, except for two treatments in late-stage Mexico, for a total of 330 experimental communities (electronic supplementary material, appendix S2). Upon retrieval, we brought communities to a laboratory and used a stereo-microscope to identify individuals to the lowest taxonomic level, often species, as a measurement of taxonomic richness, and recorded percent cover for each taxon and available open space from a 50-point grid. We quantified all living invertebrates in the communities, including those attached to the panel as well as those attached to other invertebrates (i.e. overgrowth). We sampled individuals of all morphospecies and confirmed field identifications with the help of taxonomic experts and/or genetic barcodes.

(b) . Functional trait diversity

The ability to infer biotic interactions from changes in trait diversity and structure relies on quantifying relevant traits [29]. Some sessile marine invertebrates use calcification as structural defence to deter predators [30,31], while others use colour to warn predators of unpalatable taste or chemical defences [32]. Different morphologies mediate the acquisition of resources, such as food, space and oxygen [33,34]. Reproductive traits can shift in response to a change in consumer pressure and competition [35,36]. We, therefore, characterized the functional space of 179 taxa (genus/species 149; family/class 30) with traits associated with competitive abilities, defence mechanisms, and reproduction (table 1, as published in [5]). Taxa-specific traits were categorical, binary, discrete or continuous and were collected through field measurements, observations or from literature, and when unavailable, we used values from the closest taxonomic level. For field measurements of colour, defence, growth form, organic and water content, we collected an average of five individuals per taxa for photographic reference and estimation of wet, dry and ash-free dry weight [5]. We calculated water content [54] and organic content [55] per sample and used mean trait values per taxa to calculate functional indices [23]. The taxonomic richness of each community was converted into a presence–absence matrix per site, and together with the trait matrix, we calculated functional diversity using the similarity index, Rao's quadratic entropy (RaoQ). RaoQ indicates the degree of trait similarity of communities with lower values indicating higher trait similarity and higher values greater trait dissimilarity [56].

Table 1.

List of traits considered in this study with their surrogate function and corresponding reference, as published in [5].

trait	category	data type	source	function	reference
organic content		continuous	field/laboratory	palatability, growth	[37,38]
water content		continuous	field/laboratory
colour	bright, dull, dark, transparent, white	categorical	field/laboratory	defence	[39,40]
structural defence	calcified structure, uncalcified structure, no structure	categorical	field/observation		[30,31,41]
sociability	colonial, solitary	binary	literature	competition, defence, resource acquisition	[42]
growth form	encrusting, erect, arborescent, massive, runner, stolonate	categorical	literature/observation	resource acquisition, competition, growth	[43]
feeding structure	with feeding appendages, without feeding appendages	binary	field/observation	competition, defence, resource acquisition	[44–48]
asexual reproduction	yes/no	binary	literature	competition, resource acquisition, colonization	[49]
sexual reproduction	hermaphroditic, gonochoristic, simultaneous	categorical	literature		[50]
larval duration (max hours)		continuous	literature	dispersal, competition	[51]
egg size (µm)		continuous	literature	competition, colonization/dispersal, predation	[52,53]
no. eggs per individual		discrete	literature
larval development	simultaneous, lecithotrophic, planktotrophic	categorical	literature
fertilization type	oviparous, ovoviviparous, viviparous	categorical	literature

(c) . Functional structure

Community weighed means (CWM) were used to assess the functional structure of communities. CWMs from categorical traits become percentages of trait affinity for each category, while the percent affinity from binary traits become mutually exclusive and only one of the two categories were kept for analyses (table 1). Continuous or discrete traits represent the mean value weighted by their abundance. Prior to analysis, we standardized CWMs to values ranging from 0–100 by obtaining proportions out of the maximum mean trait value and multiplied by 100. For each region, we generated a community × CWM matrix with 34 corresponding traits for subsequent analysis.

(d) . Statistical analyses

To explore the effect of treatment and assembly time on the observed functional diversity (RaoQ) of communities across latitude, we built a linear mixed model (LMM) with region, assembly time, treatment and their interaction as fixed factors and corresponding random factors of site nested within region interacting with the fixed components. We tested for regional variation of functional diversity with groups identified a priori using planned contrasts comparing RaoQ from predation treatments as well as each assembly time regardless of predation treatments by region. For this analysis, we excluded Mexico for a balanced statistical design and focused our treatment comparisons between the ambient (i.e. open) and control (i.e. partial cage) against the reduced predation (i.e. caged) treatment.

To explore non-random trait patterns, we compared the observed functional diversity (RaoQ) to null expectations with a null model designed to test for biotic effects on assembly. This approach requires (1) constraining the randomizations at a spatial scale where abiotic heterogeneity is minimized, and (2) defining the species pool independently for each site while including potential colonizers or ‘dark diversity’ [8,57]. We, therefore, randomized the community matrix constraining the richness of each community and the frequency of all species per site (i.e. 12 sites), and included the species pool from both assembly stages to include species absent during sampling but potentially able to colonize communities [8,57]. For each community, we re-sampled the species matrix 999 times with the independentswap algorithm from the picante R package v. 1.8.2 [58] and re-calculated RaoQ. We then compared the observed versus the simulated functional diversity with standardized effect size (SES) as follows:

$$\mathrm{SES}\hspace{0.17em}={\mathrm{FD}}_{\mathrm{observed}}\hspace{0.17em}\text{–}\hspace{0.17em}\mathrm{F}{\mathrm{D}}_{\mathrm{mean}(\mathrm{random})}/\hspace{0.17em}\mathrm{F}{\mathrm{D}}_{\mathrm{sd}\hspace{0.17em}(\mathrm{random})}$$

[8],

and used Wilcoxon signed-rank tests to find significant departure from zero (i.e. random) of SES values pooled by region [59]. Significant negative SES values indicate non-random trait convergence while significant positive values indicate non-random trait divergence. RaoQ is a preferred index for detecting assembly as randomization tests of these values are robust to differences in species richness [60]. In the communities studied here, regional differences in richness were observed, with the highest values found in subtropical Mexico [27]. We performed all statistical analyses including all three treatments (i.e. caged, open and cage control). For Mexico, we performed a second round of randomizations that excluded the 12-month communities as we only had available the ambient predation treatment for this late assembly stage. We found similar random patterns for all three treatments during early assembly in Mexico and show only one set of results (12-month communities included) (figure 2).

Figure 2.

The distributions of standard effect size (SES) for all regions with three sites pooled for each region (high latitude: blue, low latitude: green), treatment (i.e. reduced predation, ambient predation, and cage control), and assembly stage (i.e. 3 and 12-month) shown with box plots. Lines within boxes are medians, box ends are quartiles, whiskers extend to values no larger than 1.5 times quartiles, and dots represent outliers. Non-random trait distributions correspond to solid coloured box plots, random distributions are white box plots. SES significantly greater or smaller than zero are based on Wilcoxon signed-rank tests. *** p < 0.001; ** p < 0.01; * p < 0.05. (Online version in colour.)

To examine changes in space availability as a proxy for resource limitation over time, we tested for the effect of assembly stage on open space across regions using LMMs and generalized linear mixed models (GLMMs). Models included fixed effects of region, assembly stage, treatment and their interaction as fixed factors and corresponding random factors of site nested within region interacting with the fixed components. We tested regional variation of space availability using planned contrasts comparing the abundance of open space between assembly stages. First, we fitted LMMs using the abundance of open space and with an arcsine square root transformation, followed by GLMMs with binomial and beta-binomial distributions. We selected the beta-binomial GLMM as the best model based on AICc and inspection of residual diagnostics of normality, homogeneity of variances and overdispersion tests [27].

To explore shifts in trait structure, specifically CWMs, from predation and assembly stage, we used a multivariate generalized linear model (manyglm) with region, assembly stage, treatment and their interactions as predictors while accounting for correlations among traits. We found the negative-binomial distribution as most appropriate and checked assumptions of normality and homoscedasticity of residuals. Predictor significance was tested with likelihood ratio tests (LRTs) from a probability integral transform (PIT-trap) resampling which preserves the correlation structure of the multivariate response variables blocked by site and with 999 iterations [61]. For this analysis, we used the community × CWM matrix but kept only one of the two categories from binary traits (i.e. colonial, with feeding appendages, and asexual reproduction) (table 1). Pairwise comparisons were used to examine the significance of interactions of interest (i.e. region × assembly stage and region × assembly stage × treatment) using LRT. Model results include tests for the whole CWMs matrix and univariate tests for each CWM. Trait univariate p-values were adjusted for multiple testing with Holm's step-down procedure [62]. For this analysis, we only assessed regions with non-random trait distributions (i.e. significant SES values) as biotic interactions likely influenced trait dynamics (i.e. Alaska, California and Panama). From the manyglm univariate tests, we found many significant CWMs from the region × assembly stage and region × assembly stage × treatment interactions, therefore, we applied PCAs on reduced CWM matrices selecting only significant traits. We applied an ‘assembly stage PCA’ to visualize the differentiation in trait space among regions and assembly stages. Additionally, ‘predation PCAs’ were built to visualize how traits contributed to differentiation between predation treatments in regions where predation influenced the functional structure of communities.

To isolate patterns primarily driven by biotic filters, null model randomizations should be constrained within habitats with low abiotic heterogeneity [57]. We evaluated environmental heterogeneity across space and time for up to 1 year by calculating the coefficient of variation (CV) for temperature and salinity within sites from four opposite corners, among sites within regions, and among all regions. All statistical analyses were completed in R v. 4.2.0 [63]. Functional diversity indices were built with the FD package v. 1.0–12.1 [64], LMMs were evaluated with the lme4 package v. 1.1–30 [65], GLMMs were evaluated with the glmmTMB package v. 1.1.4 [66], estimated means were completed with the emmeans package v. 1.7.5 [67], manyglm was evaluated with the mvabund package v. 4.2.1 [68], and PCAs were evaluated with the FactoMineR package v. 2.4 [69].

3. Results

Assembly stage shaped functional diversity (RaoQ) more strongly at higher latitude regions. Lower functional diversity occurred at early versus late assembly regardless of predation treatments, with the strongest effect occurring in Alaska and California (figure 1). Contrary to our prediction, predation did not shape the observed functional diversity (RaoQ) in any region (electronic supplementary material, appendix S3). Further, during early assembly, communities in Alaska and California showed non-random trait convergence (i.e. trait similarity; negative SES values), while later in assembly, these higher latitude communities showed non-random trait divergence (i.e. trait dissimilarity; positive SES values) (figure 2). Accordingly, at these higher latitude regions, the availability of open space decreased from early to late assembly stages (figure 3). In tropical Panama during early assembly, random patterns occurred in all treatments. During late assembly, however, reduced predation resulted in trait divergence, but random patterns emerged under ambient predation (figure 2). Moreover, in Mexico, all communities at both assembly stages were randomly structured (figure 2). Detection of biotic filters is improved by constraining null models with communities from habitats with low abiotic heterogeneity [57]. Within most sites, temperature and salinity variation was negligible and at least three orders of magnitude lower than among regions (electronic supplementary material, appendix S4).

Figure 1.

Effect of assembly stage on observed functional diversity defined with RaoQ and shown as estimated marginal means (± SE) of the region × assembly time interaction from the linear mixed model (N = 268) with the following results: region, F_2,6 = 13.952, p = 0.006; assembly stage, F_1,6 = 41.069, p < 0.001; region × assembly stage, F_2,6 = 6.510, p = 0.03. All other fixed factors p > 0.05. Planned contrasts *** p = 0.001; *p < 0.05. (Online version in colour.)

Figure 3.

Effect of assembly stage on the abundance of available space and shown as estimated marginal means (± SE) of the region × assembly stage interaction from the linear mixed model with the following results: region, χ² = 11.1, p = 0.004; assembly stage, χ² = 26.5, p < 0.001; region × assembly stage, χ² = 39.1, p < 0.001; region × treatment, χ = 11.7, p = 0.02; region × assembly stage × treatment, χ² = 10, p = 0.04. Planned contrasts *** p < 0.001. (Online version in colour.)

Assembly stage shaped functional structure (CWMs) in Alaska (LRT = 226, p = 0.005), California (LRT = 838, p = 0.005) and Panama (LRT = 243, p = 0.005), while predation influenced the functional structure solely in Panama during both assembly stages (early assembly full cage versus open: LRT = 378, p = 0.005 and full cage versus partial: LRT = 348, p = 0.005; late assembly full cage versus open: LRT = 183, p = 0.005 and full cage versus partial: LRT = 205, p = 0.005; both assembly stages partial cage versus open: p > 0.05). A caging artefact was observed in Alaska during late assembly (full cage versus open: LRT = 176, p = 0.005, partial cage versus open: LRT = 127, p = 0.02, full cage versus partial: p > 0.05) perhaps as bivalves often recruited on caging material and were removed during maintenance. Univariate test results from the region × assembly stage and region × assembly stage × treatment interactions are provided in electronic supplementary material, appendix S5.

The first two axes of the assembly stage PCA explained 26.1% and 20.6% of the variation. Trait strategies in the first axis showed a continuum from colonization to competitive strategies between early and late assembly stages. This first axis separated species with lecithotrophic development and without a structure for hiding or defence, from competitors that have feeding appendages and a larval development capable of feeding (absolute loading ≥0.3). The second axis separated competitive traits between higher latitude regions during late assembly (Alaska and California). Species with an erect growth form were favoured in Alaska, while in California species with colonial growth were more common (absolute loading ≥0.3) (figure 4).

Figure 4.

Assembly stage principal component analysis (PCA) of selected species traits from significant community weighted means from the region × assembly stage manyglm univariate tests results (electronic supplementary material, appendix S5). Open shapes correspond to early assembly and filled shapes correspond to late assembly. Trait loadings with an absolute value ≥ 0.3 were kept for visualization. (Online version in colour.)

In the predation PCAs from Panama, at both assembly stages, the first two axes separated trait strategies for survival. The first axis separated calcification as a survival strategy mainly during early assembly, while the second axis showed lack of coloration (i.e. white) as a dominant trait under relaxed predation (absolute loading ≥0.3) (figure 5a,b). Additionally, during early assembly the first axis also indicated water content as a trait favoured under reduced predation (figure 5a).

Figure 5.

Panama predation principal component analysis (PCA) of selected species traits from significant community weighted means from the region × assembly stage × treatment manyglm univariate tests results (electronic supplementary material, appendix S5). Trait loadings with an absolute value ≥ 0.3 were kept for visualization. (a) Early assembly (3-month). (b) Late assembly (12-month). (Online version in colour.)

4. Discussion

Traits of near-shore communities across regions from multiple latitudes responded differently to assembly stage and predator exclusion, suggesting distinct biotic interactions influence community assembly. Trait divergence and an increase in traits associated with competitive abilities were observed at later assembly stages at higher latitude regions when competition is expected to be stronger, and multiple trait trade-offs were also evident between early and late assembly at these regions. Predation influenced the trait structure in tropical Panama during both assembly stages, and calcification as structural defence was more common with ambient predation during early assembly. Shifts in the trait structure of communities confirmed competition becomes a structuring force during late assembly at higher latitude regions and predation is most relevant in the tropics.

At higher latitude regions, a shift to trait divergence was observed with later assembly, as space, an important resource in this system, became limiting and competition likely intensified [8]. In Panama, under relaxed predation and during late assembly, trait divergence was also observed, however, an association with competitive traits was not evident. Competitive dynamics often gain relevance later in development [12], and strong competitors can dominate in late successional stages in sessile communities [70]. In Alaska and California at late assembly, slower settlers (i.e. those with feeding larvae and longer larval duration) were more common, but dominant growth forms (i.e. colonial or erect) differed between the two regions. Planktotrophic (feeding) larvae are capable of re-settling to avoid non-optimal substrate [71], colonial forms are associated with high overgrowth abilities [72], and erect forms can provide better feeding performance [73]. Therefore, trait divergence observed later in development likely resulted from competitive dynamics and a mix of common and distinct trait mechanisms that define competitive abilities at each region.

While we observed trait divergence at later community assembly, we also observed trait convergence at early assembly at higher latitude regions. Competitive exclusion may lead to trait similarity or convergence [18], however, we suspect competitive exclusion was unlikely, since space was readily available at the early assembly stage. Instead, colonizers perhaps from nearby source communities and with faster establishment strategies (i.e. lecithotrophic development and shorter larval duration), but that could be competitively inferior dominated the communities and limited trait diversity early in assembly. Therefore, trait convergence as a result of colonization dynamics at early assembly and divergence from competition at later assembly would suggest that competition–colonization trade-offs [74] are a likely mechanism shaping community assembly over time at higher latitude regions.

In Panama, we also observed trait divergence at later assembly, but only under relaxed predation. In addition, the functional structure of communities from Panama shifted with predation treatments, and traits that optimized survival included calcified structures that provide protection [31]. Moreover, low water content decreases palatability to discourage predation [5] and colour or lack thereof defines prey vulnerability [39], and these traits differentiated communities that assembled under ambient and relaxed predation at both assembly stages. While consumer pressure can cause non-random trait patterns, other factors that interact with consumer pressure, such as interaction history, can cause trait divergence. In a parallel study, we found predation increased the functional diversity of introduced species in these early assembly communities in Panama [5]. Together, our results suggest that predation may have different functional effects on subsets of taxa, but both studies highlight the importance of predation shaping complementary elements of functional diversity and structure in Panama during both assembly stages [72,75]. Overall, both predation and assembly stages influenced the functional trait structure in Panama suggesting predation-shaped communities at both assembly stages while competition could become more relevant at later stages of community development in the absence of predation.

In Mexico, random patterns were observed at both assembly stages and may result from weaker species interactions than at other latitudes. Random assembly is expected when resources are widely available, species interactions are weak to non-existent, or random colonization prevails [12,59]. Moreover, a shift away from random assembly may occur through time as dispersal processes can influence early stages of assembly and successive filtering from biotic interactions dominate as communities mature [11]. In subtropical regions, predation can have an intermediate effect on prey when compared to higher and lower latitudes [4,27], weakening any potential for non-random trait patterns or shifts in functional structure due to predation. Subtropical regions may therefore be transition zones between tropical and temperate dynamics as weaker but detectable predation effects have been observed [25,27], particularly at early assembly.

Temporal shifts help disentangle the complex dynamics that govern coexistence across spatial [76], environmental [75], and in this study multiple regions across latitude. By focusing on small-scale dynamics, we observed how species interactions influence the distribution of traits across time and latitude. At higher latitude regions, a synchronous change from trait convergence to divergence and colonizing to competitive traits between assembly stages provides strong support for competition–colonization trade-offs defining assembly over a 1-year timescale. In the tropical region, during both assembly stages, however, strong predation resulted in pervasiveness of predation-resistant traits, providing additional evidence that predation has a greater influence on community dynamics at tropical latitudes [2,4]. The results from our regional comparisons align with evidence from other studies showing predation is stronger in the tropics [10,62,77], and the stage of community development determines the importance of competition as a structuring force [12]. Community assembly processes continue to be a topic of debate and trait-based assembly explored at local scales and through time provides a mechanistic tool to understand community structure across biogeographical gradients. This study is the first to employ a large-scale experimental approach to demonstrate evidence that biotic interactions can shift trait strategies of communities over time, with outcomes hinging on geographical location.

Ethics

This research was conducted under permit no. CF-16–016 from the State of Alaska Department of Fish and Game, United States, permit no. SC-4765 from the State of California Department of Fish and Wildlife, United States, permit no. PPF/DGOPA-291/17 from the Dirección General de Ordenamiento Pesquero y Acuicola, Mexico, and permit no. SE/A-7–17 MiAmbiente, Republica de Panama.

Data accessibility

Community composition and open space data that support the findings of this study are available at https://doi.org/10.26008/1912/bco-dmo.861250.1 [78]. Richness data are available at https://doi.org/10.26008/1912/bco-dmo.861234.1 [79]. Trait data are available at https://doi.org/10.26008/1912/bco-dmo.883700.1 [80]. Code that support the findings of this study are available as electronic supplementary material [81].

Authors' contributions

D.P.L.: conceptualization, data curation, formal analysis, funding acquisition, investigation, methodology, project administration, visualization, writing—original draft, writing—review and editing; A.L.F.: conceptualization, funding acquisition, investigation, methodology, project administration, writing—review and editing.

All authors gave final approval for publication and agreed to be held accountable for the work performed therein.

Conflict of interest declaration

We declare we have no competing interests.

Funding

This research was supported by NSF OCE grant 1434528 to A.L.F., and by a Future Faculty Fellowship and dissertation completion grant from Temple University to D.P.L. Null models were conducted on TU high-performance computing resources and supported by NSF grant 1625061 and the US Army Research Laboratory contract number W911NF-16-2-0189.