Heterogeneity in plant–soil feedbacks and resident population dynamics affect mutual invasibility



  1. Understanding the mechanisms governing coexistence is a central goal in ecology and has implications for conserving and restoring communities, yet the high diversity in many plant communities is difficult to explain. Theory suggests that plant–soil feedbacks (PSF) can lead to frequency-dependent coexistence by suppressing conspecifics more than heterospecifics, potentially helping to explain high-diversity plant communities. In addition, species-specific population dynamics, including the rate at which individuals are replaced in a population or population turnover rate, may influence coexistence outcomes.
  2. We have created a rigorous test of the coexistence predictions of theory by generating a soil heterogeneity experiment in the field and testing for mutual invasibility by establishing resident populations, then experimentally invading them. Experimental tests of mutual invasibility can demonstrate coexistence because, if species are able to invade one another's populations when at low density, they should exhibit long-term coexistence. We use pairs of congeners in this experiment that coexist at small spatial scales, sometimes within cm, at our field site.
  3. We demonstrate that invader individuals established better in congener's soils than in conspecific soils, consistent with plant–soil feedback mediated coexistence. This effect was often mediated by competition with established resident plants.
  4. Further, we show that soil heterogeneity interacted with the population turnover rate of the resident population to influence invasibility (< 0.10), consistent with the theoretical prediction that a plant's population dynamics will interact with heterogeneity to influence coexistence.
  5. Synthesis. Plant–soil feedbacks (PSF) can in theory lead to frequency-dependent coexistence, and reciprocally, negative feedback effects in glasshouse experiments are often consistent with this prediction. We provide the first field test of mutual invasibility structured by PSF, demonstrating that PSF can lead to coexistence when they create a patchy, or heterogeneous, soil environment. This work suggests that understanding the influence of PSF on diversity necessitates understanding the spatial scale at which soil heterogeneity emerges in the field. Thus, high diversity might be maintained in plant communities by heterogeneity created by plants' influence on the soil, and this outcome depends strongly on population dynamics.


One of the primary goals of community ecology is to explain and predict the processes governing community assembly. However, plant communities are still frequently more diverse than predicted by coexistence theory (Agrawal et al. 2007), and restoration ecologists need to understand coexistence mechanisms to successfully restore diverse native communities (Suding 2011). Environmental heterogeneity is a theoretically important but rarely experimentally tested mechanism that can help explain high levels of community diversity (Melbourne et al. 2007). Recent work suggests that the variances in environmental drivers are as important as the means in governing invasibility patterns (Shurin et al. 2010), further suggesting that heterogeneity may be as or more important to community assembly than the mean environment.

While observational studies generally find a positive correlation between heterogeneity and diversity in plant communities, experiments manipulating soil nutrient heterogeneity generally find the opposite (Lundholm 2009; Reynolds & Haubensak 2009). Theory that incorporates heterogeneity typically assumes that heterogeneity is not affected by the community composition and is for example governed by abiotic factors such as landscape topology (e.g. Pacala and Tilman 1994, Chesson 2000a,b; reviewed in Melbourne et al. 2007). However, much of the environmental heterogeneity relevant to plant communities may be dependent on community composition (e.g. Bever, Westover & Antonovics 1997). Considering the role of this type of heterogeneity in coexistence may help explain the apparently contradictory findings in the literature. Thus, we manipulate heterogeneity resulting from plant–soil feedbacks and test the coexistence consequences of this heterogeneity (Brandt, del Pino & Burns 2014).

Soil heterogeneity includes components of the soil that result from plant–soil feedbacks (Ettema & Wardle 2002; Bever et al. 2010). Plant–soil feedbacks occur when plants influence the soil in a manner that affects subsequent plant performance in that soil (Bever et al. 2010). Plant–soil feedbacks can lead to coexistence by resulting in more negative intraspecific effects than interspecific effects (Bever, Westover & Antonovics 1997), and such effects are quite common in glasshouse experiments (Klironomos 2002; Kulmatiski, Beard & Stark 2006; Kulmatiski et al. 2008). For example, plant species often have species-specific soil pathogens that build up in their rhizosphere over time (e.g. Packer & Clay 2000), leading to negative intraspecific effects. Because plant–soil feedbacks occur in a spatial context in the field, whether such feedbacks lead to coexistence will depend on the spatial arrangement of soil patches. Although the mean effects of plant–soil feedbacks on individual plant performance are well characterized (Kulmatiski et al. 2008), the consequences of spatial heterogeneity in these feedbacks for populations and communities are largely unknown (Reynolds & Haubensak 2009; Brandt et al. 2013). Thus, to determine the coexistence consequences of plant–soil feedbacks, experiments in a spatial context in the field are needed.

The apparent conflict in the literature on environmental heterogeneity and diversity might also occur because the effects of heterogeneity could depend on species' population dynamics. While early theory suggested that increasing environmental heterogeneity should increase invasibility (reviewed in Davis, Grime & Thompson 2000; Melbourne et al. 2007), more recent theory suggests that the influence of heterogeneity on invasibility may depend on population turnover rate, or the rate at which individuals die and are replaced in the population (Schoolmaster & Snyder 2007; Snyder 2007). Population turnover rate is a function of species' life history – shorter-lived species will have higher mortality and higher fecundities resulting in higher population turnover rates than longer-lived species. Invasion biology has long predicted that the life history of the species should influence invasibility (e.g. Lonsdale 1999), and higher turnover resident populations should leave more open space for invasion. In addition, population dynamics, including population turnover rate, might also depend on environmental disturbance, where high levels of disturbance lead to high resident mortality and thus high invasibility (e.g. Burke & Grime 1996). Thus, we predict that the population turnover rate of a resident population will influence its invasibility.

We test these predictions using mutual invasibility experiments because mutual invasibility is a necessary (but not always sufficient) criterion by which theory evaluates stable coexistence (Chesson 2000b). Mutual invasibility experiments invade two species into one another's established, resident populations, evaluating whether each population can increase from low density. If two populations are mutually invasible, this provides empirically rigorous evidence for coexistence between them. Theory has made considerable progress in predicting the role of environmental heterogeneity in governing invasibility, but strong empirical tests of this theory are rare, with a recent review finding only three studies of mutual invasibility for plants and none performing mutual invasibility experiments in plant systems (Siepielski & McPeek 2010), suggesting that paired mutual invasibility studies are greatly needed to rigorously test coexistence theory in plant communities.

We paired species by relatedness because closely related species are often similar to one another (e.g. Silvertown & Dodd 1996; Peterson, Soberon & Sanchez-Cordero 1999; Burns & Strauss 2011), consistent with the high degree of morphological similarity between our species pairs (JHB, pers. obs.). Using congeneric pairs, we follow a long tradition of using congeneric species pairs to control for phylogenetic relatedness in ecological studies (Felsenstein 1985; e.g. Fiedler 1987). Each species in a congeneric pair was planted as a resident, allowed to establish, and then invaded with its congener. Each species within a pair was treated as both the resident and the invader, thus testing for mutual invasibility (i.e. coexistence) (Chesson 2000a). Pairing species in this fashion also allowed us to create a feasible experimental design, because conducting mutual invasibility experiments for all possible pairs of species in a community was outside the scope of the study.

We used this paired design to create an experimental test of coexistence, including a test for the role of plant–soil feedbacks in coexistence. We conducted a factorial experiment with two levels of environmental heterogeneity (heterogeneous soil patches, homogenized soil) and two levels of resident population turnover (ambient, high) for six perennial plant species (three congeneric pairs) (Fig. 1). Heterogeneous soil treatments contained patches of soil collected from the resident and invader congeneric pair, and homogeneous soil treatments contained patches that were a mixture of both soil types (Fig. 1). Population turnover was operationally defined as mortality followed by replacement. Manipulating resident population turnover is experimentally tractable in perennial plant systems because mortality and replacement rate can both be manipulated. We established populations of residents for each of six focal species and then experimentally invaded these populations with their congeners and monitored invader establishment, evaluating potential coexistence consequences of our experimental treatments. In this study, we address three main questions: at the scale of individual plants: (i) How do plant–soil feedbacks and resident population turnover influence individual plant establishment? To determine the potential role of competition in driving responses to treatments, we ask: (ii) Is invader establishment in response to experimental treatments mediated by competition with resident plants? Finally, to test the predictions of coexistence theory, at the scale of populations, we ask (iii) Do soil heterogeneity and resident population turnover interact to influence invasibility?

Figure 1.

The experimental design to determine the coexistence consequences of two levels of soil heterogeneity (heterogeneous and homogeneous) and two levels of resident population turnover (ambient, high = 2× ambient). Heterogeneous plots were created with three patches of resident soil and three patches of invader soil, and homogeneous treatments were a mixture of these two soil types. High resident turnover was implemented by clipping half of the established residents in a plot, chosen at random, and adding two additional seeds marked with toothpicks in those positions, to keep mortality/replacement constant (Appendix S1). Thus, we show 2× as many residents in ambient treatment plots, and the number of resident individuals per plot within a treatment in the field varied. This design was implemented for six species as invaders, with their congener as the resident, in randomized field plots. Arrows point to positions where high turnover was implemented (i.e. mortality + seed addition). Timeline indicates times when major activities in the experiment occurred. Invader establishment was measured as a binary response (0 = no plant established, 1 = plant established) at each planting position, indicated with red triangles, at the final invader census.

Materials and methods

Study System

We performed a field experiment at Case Western Reserve University's Valley Ridge Farm (‘University Farm’) (41°29′ N, 81°25′ W), Hunting Valley, OH, USA, to test the prediction that heterogeneity in plant–soil feedbacks can influence invasibility and that soil heterogeneity interacts with resident population turnover to influence invasibility. University Farm is an approximately 157 ha field site consisting primarily of old fields and beech-maple hardwood forest. We used six short-lived perennial species found at the site: Plantago lanceolata L., Plantago major L., Rumex crispus L., Rumex obtusifolius L., Solanum carolinense L. and Solanum dulcamara L. Five of these species are herbaceous, while S. dulcamara is woody. Both Solanum species have a sub-shrub growth form in this experiment (JHB, pers. obs.), and all species used here are considered weedy or invasive (USDA 2010). Here, we use these species as a test of coexistence theory, not as a test of invasiveness per se (i.e. spread rate in the landscape, sensu Richardson et al. 2000). Using pairs of closely related species, we control to some extent for differences in growth form (Felsenstein 1985). These pairs of taxa co-occur at our field site, including within cm (Rumex, Plantago) to m (Solanum) of each other.

Experimental Design

For each species pair within a genus, we manipulated spatial soil heterogeneity (homogenized, heterogeneous) and population turnover of the resident species (ambient, high) in a factorial design fully randomized in the field (Fig. 1). For each of the six resident species, we established five replicates per soil heterogeneity and population turnover treatment for a total of 120 pots. Soil treatments were established using field-collected soils, which incorporates abiotic and biotic variation in the soil environment resulting from plant–soil feedbacks (Bever et al. 2010). Soil was collected from locations where each species was abundant from the zone of root formation, which includes the rhizosphere soil and soil found within the root ball, and care was taken to prevent cross contamination of soil. Pseudoreplication of soil samples (e.g. a single rare pathogen being mixed across replicates) was minimized by mixing soil in ~ 8 subsets per species, haphazardly arranged across the replicates (see additional methods details in Brandt et al. 2013). Large (46 cm diameter × 38 cm deep) pots sunken into the ground were used to create replicate plots of each soil treatment. Each plot was divided into six patches, and each patch was 10 cm (length) × 10 cm (width) × 18 cm (deep). Each plot consisted of either homogenized soil from each species pair or a discrete mosaic of soils collected from each congener (Fig. 1). To create the heterogeneous soil treatment, soil from each species in a congeneric pair was placed into separate patches, creating a spatially heterogeneous mosaic of soil types within a plot (i.e., three patches per species per plot; Fig. 2a). To create the homogenized soil treatment, we placed equal parts of each soil type within a congeneric pair into each of six patches (for additional methods details, Brandt, del Pino & Burns 2014).

Figure 2.

(a) An example heterogeneous soil treatment plot. Soils from the soon-to-be-planted resident, Solanum carolinense, are in patches 1, 3 and 5. Soils from the invader, Solanum dulcamara, are in patches 2, 4 and 6. (b) A plot with resident seeds of S. carolinense planted on each blue toothpick. Residents were planted in the upper left and lower right corner of each patch (Fig. 1). (c) A plot with resident S. carolinense establishing, and seeds of the invader, S. dulcamara, added on red toothpicks. Invader seeds were added to the upper right and lower left corner of each patch (Fig. 1). (d) A high turnover plot with resident S. carolinense after half of the individuals have been killed and additional seeds added to each position where mortality was enforced. Numbers of established residents varied across plots, and thus, the number of individuals killed varied across plots and was always 1/2 of the established residents in the high turnover treatment plots. Red toothpicks indicate invader seeds of S. dulcamara.

For each species pair, resident populations were initiated on 30 June 2011 by planting two seeds in each of two corner positions per patch (n = 12 planting positions per plot for 24 seeds per plot). Seeds were individually marked with a coloured toothpick (Fig. 2b). Thus, in establishing the resident population, each plot contained seeds from only one species (Fig. 2b). Seeds were locally collected from naturalized populations at University Farm in the fall the year before planting from ~10 maternal plants ~5 m apart. Resident populations were generally well-established with many large and reproductive resident plants, creating a clear establishment disadvantage for invaders (JHB, pers. obs.). After allowing resident species to grow for approximately 3 months, seeds of the congener species (invader seeds) were added to each plot on 11 October 2011 (Fig. 2c). For each patch within a plot, two seeds of the congener species were added in the remaining two corner positions, for a total of 12 planting positions and 24 seeds per plot (Fig. 2c).

Population turnover treatments (i.e. ambient and high turnover) on resident plants were applied on 9 July 2012. Resident plants in plots assigned to the ambient turnover treatment were not altered, but natural mortality was allowed and monitored. High turnover treatments were established by inducing mortality of half of the individuals per plot, which were chosen at random (Fig. 2d). Thus, different numbers of plants were killed in different plots, depending on the number of residents established in that plot at the time the treatment was applied. To minimize soil disturbance differences between turnover treatments, individuals chosen for the high turnover treatment were clipped at the base and the remaining plant material was painted with glyphosate (undiluted; Razor® Pro, McClure, PA, USA) to induce root mortality. Three weeks after induced mortality, two additional seeds of the resident species were added to locations where mortality was induced to maintain mortality/replacement between turnover treatments (Fig. 1, Appendix S1 in Supporting Information).

To ensure establishment and allow populations to turn over, we added additional seeds of the resident species on five different dates (27 July 2011, 16 August 2011, 6 October 2011, 11 June 2012 and 31 July 2012) to each position without a plant. Two invader seeds per position without a plant were also added on 12 June 2012 and 15 August 2012. Thus, this experimental design minimizes seed limitation as a factor influencing establishment (Clark et al. 2007). We measured invader establishment at a planting position as a categorical response on 24 September 2012 (Fig. 1).


Schoolmaster & Snyder (2007) predict that species coexistence can be influenced by the interaction of resident population turnover rate and the rate of spatiotemporal variation in patch quality. One mechanism potentially driving the spatiotemporal variation in patch quality is plant–soil feedbacks (Bever, Westover & Antonovics 1997). Here, our analyses test (i) the prediction that plant–soil feedbacks can influence coexistence, such that negative plant–soil feedbacks will enhance plant's establishment in heterospecific soils compared with conspecific soils (Bever, Westover & Antonovics 1997), using individual-level analyses (below). (ii) Secondly, we test the theoretical prediction that mutual invasibility should depend on environmental heterogeneity interacting with resident population turnover rate (Schoolmaster & Snyder 2007). Because invasibility is population growth rate from low density, we test the predictions of this model using population-level analyses (below). Further, the predictions of this model were generated under the following assumption (Test of theoretical model assumptions, below), which we test for our data.

Test of theoretical model assumptions

To test the assumption of theoretical models that mortality/replacement rates, ‘δ:F’ in Schoolmaster & Snyder (2007), did not differ among treatments, we calculated this ratio at each position and conducted a mixed effects model including plot as a random effect and treatment and interactions as fixed effects on square-root-transformed mortality/replacement (details in Appendix S1).

Individual Scale

(1) How do plant–soil feedbacks and resident population turnover influence individual plant establishment?

To determine whether individual invader establishment was a function of soil patch type interacting with resident population turnover rate, we conducted a logistic regression on whether planted invaders had established at a given position within a plot by the end of the 2012 field season (24 September 2012). Invader establishment was modelled as a binary response variable, using a generalized linear model using the glm function with binomial error structure and invader identity (of six invader species), soil type (homogeneous, resident = congener, invader = conspecific) and resident turnover treatment as main fixed effects, all possible interactions among those effects, and the number of invader seeds added as a covariate. Thus, this model was Y{i,j,k} ~ Binom(n{i,j,k}, p {i,j,k}) where Y = {0,1}, n is the number of seeds added to position i, plot j, treatment k and p{i,j,k} is a function of soil type, turnover treatment and invader identity. Sample size was 1440 positions across all plots with five plots per treatment combination and 12 invader positions per plot (i.e. 5 × 12 = 60 samples per treatment level).

Next, we conducted independent contrasts to test specific a priori hypotheses suggested by theory. To contrast the average influence of soil heterogeneity on individual invader establishment within the ambient turnover treatment, we again conducted a logistic regression using a type III SS glm in R (R Development Core Team 2008) on invader as a binary response at the end of the season (24 September 2012), including the number of seeds added to that position as a covariate for only the ambient turnover treatment (sample size = 720 individual positions). We used contrasts to test for effects of soil type (resident, invader) within the heterogeneous treatment and for differences between soil treatments (homogeneous, heterogeneous). If plant–soil feedbacks are primarily negative, such that they could enhance coexistence, we predict that invader establishment will be higher in congener's soils than in conspecific soil (resident > invader). If these feedbacks depend on the patchiness of the soil, then we expect that there will be differences in invader establishment between homogeneous and heterogeneous soils.

(2) Does competition with residents mediate treatment effects on invader establishment?

Individual invader establishment might be mediated by multiple mechanisms in this experiment. Direct effects of soil patch quality could influence establishment, for example via plant–soil feedbacks. In addition, indirect effects of competition with resident plants, whose establishment was influenced by soil treatment (Brandt et al. 2013), could also influence an invader individual's establishment. To determine to what extent invader establishment might be explained by patch quality and competition with residents, we conducted a mediator analysis at the position scale (Baron & Kenny 1986). Mediator analysis partitions the relative strength of direct and indirect (mediator) effects, allowing us to ask whether the effect of soil treatment was mediated by resident plant establishment (indirect effect), directly influenced by treatment, or a combination of both. The presence of neighbouring residents may also be influenced by turnover treatment, where fewer resident neighbours may occur in high turnover treatments. To partition the direct and indirect effects, we conducted the same logistic regression analysis as above (Question 1) for each species with and without including a potential mediator, the presence of resident neighbours, evaluated on 24 September 2012, coded as a factor. Alternative analyses using the number of resident neighbours as a continuous predictor resulted in similar conclusions and are not shown. We tested for interactions between neighbour presence and treatment effects and dropped these interactions if they were not significant (> 0.10). If the treatment effects decline with the addition of resident competition to the model, we interpret this as evidence that competition with residents mediates the treatment effect. For example, high resident establishment in a congener's (i.e. invader's) soil patches could result in strong competition in those patches and thus low invader establishment in its own soil. Thus, plant–soil feedbacks could indirectly influence invader establishment, through their effects on resident establishment. If adding resident competition results in no change in treatment effects, then competition with residents does not mediate the treatment effects. For example, soil patch type could directly influence the establishment of invaders through plant–soil feedbacks.

Population Scale

(3) Is invasibility a function of soil heterogeneity interacting with resident population turnover?

Theory predicts that environmental heterogeneity should interact with resident population turnover to influence invasibility (Schoolmaster & Snyder 2007). To determine treatment effects on invasibility, which is a population scale response, we asked what proportion of invader individuals planted established (by 24 September 2012) per plot. We divided the number of invaders in the plot by the number of seeds added, which differed among plots because we continued to add seeds to positions with no establishment (see Experimental Design above). The data were zero-inflated (i.e. a large number of plots had zero invaders) and could not be transformed to normality. Next, we transformed the proportion invader establishment to count data by multiplying by 1000 to retain the variance and truncating the data to integer values, so that we could run a zero-inflated model, which requires data with integer values. Finally, we tested each component in the maximal zero-inflated model that we had the power to fit (corrected number of invaders ~ Invader species + Soil treatment × Turnover treatment | Invader species + Soil treatment × Turnover treatment). These models contain two components: the count component (on the left of |) and the zero-inflated component (on the right of the |). The count component was modelled with a negative binomial distribution, and the zero-inflated component was modelled with a binary distribution. The zero-inflated component tests the hypothesis that the presence of invaders was a function of invader species identity and treatment. Further, the count component tests the hypothesis that the number of invaders in each plot, given the presence of invaders, was a function of invader identity and treatment. Theory predicts an interaction between soil heterogeneity and resident population turnover on invasibility, and this model tests for that interaction on two components of invasion, establishment and population size at the plot scale. These analyses used the zeroinfl function in the pscl library (Zeileis, Kleiber & Jackman 2008).

All analyses were conducted in R (R Development Core Team 2008), and diagnostic plots were examined to test model assumptions. Data used in these analyses are archived at the Dryad Digital Repository (Burns & Brandt 2014).


Test of theoretical model assumptions

The ratio of mortality to replacement was not a function of treatments (Appendix S1), suggesting that turnover treatment was not confounded with differences in mortality/replacement. Observed turnover rates for the residents at the plot scale were generally low and were somewhat higher for the high turnover treatment (ambient = 5.88 ± 0.69 SE, high = 6.33 ± 0.66 SE individuals per plot).

Individual scale

(1) How do plant–soil feedbacks and resident population turnover influence individual plant establishment?

Soil patch type interacted with population turnover rate to influence individual invader establishment (Table 1). Successful establishment of an invader individual depended upon a marginally significant three-way interaction among invader identity, soil type and resident population turnover (Table 1). On average, invader individuals were marginally more likely to establish when soil was heterogeneous than when soil was homogeneous, at ambient resident turnover (proportion of seeds established per position: hetero = 0.048 ± 0.006 SE, homo = 0.046 ± 0.006 SE, z = −1.683, = 0.09). In addition, soil patch type within the heterogeneous treatment influenced invader establishment, where invaders had higher establishment in their congener's soil than in conspecific soil (resident soil = 0.059 ± 0.009 SE, invader soil = 0.038 ± 0.007 SE, z = −1.988, = 0.05), and there was no interaction between soil type and invader species identity (> 0.25).

Table 1. Individual invader establishment (binary response) at a given position was a function of the number of invader seeds added and the interaction among invader identity, soil type and resident population turnover treatment
  1. Soil types were homogeneous soil, resident (congener) soil and invader (conspecific) soil. Logistic regression with model deviance = 662.3, model AIC = 1204.7. Effects with < 0.05 were interpreted as significant and shown in bold. Effects with < 0.10 were interpreted as marginally significant and shown in italics.

Invader seeds added 1 936.4 1476.7 274.10 < 0.001
Invader identity 5 686.1 1218.5 23.82 < 0.001
Soil type2663.31201.60.95> 0.25
Turnover treatment1662.41202.70.05> 0.25
Invader × soil10672.21194.69.92> 0.25
Invader × turnover 5 676.3 1208.7 14.04 0.02
Soil × turnover2665.11203.52.81> 0.10
Invader × soil × turnover 10 678.8 1201.1 16.47 0.09

(2) Does competition with residents mediate the effect of treatments on invader establishment?

Competition with established residents mediated the effects of experimental treatments for three of six experimental species: P. lanceolata, S. carolinense and S. dulcamara (Fig. 3 and Table S2). Plantago lanceolata invader establishment was often reduced in the presence of resident neighbours, suggesting that the high turnover treatment often reduced competition and increased invader success (Fig. S2). Further, the effect of competition depended on soil type and turnover treatment, such that P. lanceolata invaders established better in conspecific than in congener soil, but only when turnover was ambient and neighbours were present and when turnover was high and neighbours were absent (Fig. S2). Solanum carolinense invader establishment was higher in the high turnover treatment than in the ambient treatment (ambient = 0.017 ± 0.007 SE, high = 0.031 ± 0.007 SE), and this effect was entirely mediated by competition with residents, which was reduced at high turnover (Table S2). Solanum dulcamara invader establishment was also mediated by competition with residents, such that establishment was lower when resident neighbours were present for both turnover treatments in homogeneous plots, but only for ambient turnover in heterogeneous plots (Fig. S2). Rumex crispus invader establishment was greater at high turnover treatments (ambient = 0.019 ± 0.007 SE, high = 0.035 ± 0.007 SE), and this effect was not mediated by resident competition (Fig. 3; Table S2). Rumex obtusifolius invader establishment was greater in homogeneous plots and invader (conspecific) soil patches than in resident (congener) soil patches (homogeneous = 0.076 ± 0.013 SE, invader soil = 0.076 ± 0.016 SE, resident soil = 0.065 ± 0.012 SE), and this effect was not mediated by resident competition (Fig. 3 and Table S2). Further, R. obtusifolius invader establishment depended on soil patch type interacting with turnover (Table S2), such that invaders established better in the high turnover treatment than the ambient treatment in invader soil and less well in the high turnover treatment than the ambient treatment in resident soil (invader soil: ambient = 0.050 ± 0.016 SE, high = 0.103 ± 0.026 SE; resident soil: ambient = 0.097 ± 0.017 SE, high = 0.033 ± 0.014 SE). Finally, P. major was a universally poor invader in all plots, and its invasion was not influenced by treatment or competition with residents (Fig. 3 and Table S2).

Figure 3.

Summary of the mediator analysis, testing for direct effects of experimental treatments on invader establishment and the possible indirect mediation of these effects through resident establishment (Table S2). Solid lines indicate relationships where treatment effects were significant in the model without the mediator (Table S2). The soil × turnover interaction did not significantly influence invader establishment for Plantago lanceolata and Solanum dulcamara when competition with resident neighbours was removed from the model (Table S2); thus, this effect was only detectable when the putative ‘mediator’ was taken into account (dashed lines). Arrows indicate significant relationships (< 0.05) in the mediator analysis.

Population Scale

(3) Is invasibility a function of soil heterogeneity interacting with resident population turnover?

Invasibility is a population-level response, thus we conducted a plot scale analysis to determine the role of experimental treatments on invader population establishment. There was a marginally significant interaction between soil heterogeneity treatment and resident population turnover treatment on the number of invaders establishing (Table 2). On average, at ambient turnover, there was no effect of soil treatment (Fig. 4, P > 0.25), and in the high turnover treatment, there was a marginally significant trend for greater invader numbers in homogeneous plots (Fig. 4, z = 1.84, = 0.07). Under the ambient turnover treatment, three of six species had more invaders establish in the heterogeneous treatment than in the homogeneous treatment, after correcting for the number of seeds added to each plot (Table 3). The opposite was true in the high turnover treatment, where four of six species had more invaders establishing in homogeneous plots than in heterogeneous plots (Table 3), consistent with an interaction between soil heterogeneity and turnover treatment.

Table 2. Zero-inflated model for the count of the number of invader individuals per plot, a population scale analysis, as a function of invader identity and soil heterogeneity and resident population turnover treatments
  1. A zero-inflated model was conducted because many plots contained no invaders; thus, the data contained a disproportionate number of zeros. The number of invader individuals per plot was corrected for the number of seeds added to each plot (see Methods: Analyses). Effects with P < 0.05 were interpreted as significant and shown in bold. Effects with P < 0.10 were interpreted as marginally significant and shown in italics.

Invader 25.78 < 0.001 51.39 < 0.001
Soil heterogeneity treatment0.60> 0.250.24> 0.25
Turnover 3.31 0.07 0.24> 0.25
Soil × turnover 2.72 0.10 0.26> 0.25
Table 3. Mean corrected establishment proportion per plot, a population scale response, by species and treatment (mean ± SE)
  1. To correct for the number of seeds added to each plot, the total number of invader individuals was divided by the number of seeds added to the plot.

Plantago lanceolata 0.088 (0.011)0.062 (0.013)0.059 (0.021)0.119 (0.028)
Plantago major 0 (0)0.003 (0.003)0.006 (0.006)0 (0)
Rumex crispus 0.014 (0.005)0.013 (0.010)0.026 (0.009)0.040 (0.015)
Rumex obtusifolius 0.072 (0.024)0.044 (0.020)0.063 (0.016)0.071 (0.014)
Solanum carolinense 0.009 (0.006)0.012 (0.006)0.033 (0.027)0.042 (0.042)
Solanum dulcamara 0.046 (0.015)0.054 (0.028)0.042 (0.012)0.034 (0.016)
Figure 4.

The proportion of invader seeds that established in a plot (invader plants established at the end of the 2012 field season/seeds planted), a population scale response, was a function of the interaction between soil heterogeneity and resident population turnover treatments (Table 2). Means ± 1 SE.


Plant–soil feedback theory suggests that reciprocal negative feedbacks should lead to coexistence in the field (Bever, Westover & Antonovics 1997). Here, we show for the first time that soil patches in the field resulting from plant–soil feedbacks (Bever et al. 2010) influenced invader individual establishment into paired resident species populations. This suggests that the heterogeneity created by plants in the community could help explain high levels of plant coexistence at small spatial scales. Further, most paired species were able to establish in one another's populations under at least some conditions (except for P. major), consistent with stable coexistence mediated by plant–soil feedbacks, as suggested by prior theory (Bever et al. 2010). However, this individual establishment result did not scale up to the population level for all species, either suggesting a lack of statistical power at the population scale or scale dependence in this result. Further work will be needed to determine the extent to which coexistence mediated by plant–soil feedbacks is dependent on level of organization (individuals, populations).

We found evidence that competition with residents mediated treatment effects for some species (Fig. 3), consistent with a large body of literature on community assembly mechanisms (reviewed in Emerson & Gillespie 2008; Vamosi et al. 2009). Our focal species are all considered weedy, or undesirable species with high population growth and spread rates, by sources such as the Global Compendium of Weeds (GCW 2007); however, there is considerable variation in degree of weediness among species. In general, the weedier species within a congeneric pair were more limited by competition and less weedy species were more limited by patch quality, consistent with the weediest species being disturbance facilitated (e.g. Burke & Grime 1996). For example, the most weedy species here, S. dulcamara (GCW 2007), is most limited by competition, whereas the less weedy R. obtusifolius (GCW 2007) is solely limited by patch quality. Thus, the species-specific nature of these responses may become more predictable based on species' life history, although tests with larger numbers of species are necessary to assess the generality of this hypothesis.

Environmental heterogeneity has long been suggested to drive invasibility and coexistence (Davis, Grime & Thompson 2000; Amarasekare 2003; Melbourne et al. 2007), for example, by increasing the number of available niches (Elton 1958; but see Levine & D'Antonio 1999) or system-wide productivity (Davies et al. 2007). Theory predicts that environmental heterogeneity should lead to high levels of community diversity (i.e. coexistence) under some conditions (Melbourne et al. 2007). Observational studies have generally found evidence consistent with this theory, where high heterogeneity positively correlates with high community diversity (e.g. Davies et al. 2005). However, experimental tests manipulating soil nutrient heterogeneity have generally found evidence that contradicts this hypothesis, where high nutrient heterogeneity leads to lower plant community diversity (reviewed in Reynolds & Haubensak 2009), perhaps because clonal plants are able to integrate across high nutrient availability patches. We found a marginally significant positive effect of heterogeneity on invader establishment at the scale of individual plants, consistent with the theoretical prediction. These results are consistent with experimental tests manipulating soil texture, which found higher heterogeneity led to greater plant community diversity (Williams & Houseman 2013). If heterogeneity in the soil is plant-induced, perhaps influenced by factors such as soil pathogens or soil texture, this might increase coexistence to a greater degree than nutrient heterogeneity.

We dispersed seeds uniformly across our experimental plots (Figs 1 and 2), meaning that effects of heterogeneity on plant establishment do not require differences in dispersal ability among species. This is consistent with theory on coexistence via the storage effect (Chesson 2000b), which does not require differences among species in dispersal. Individuals can be ‘stored’ in patches with few species-specific pathogens or patches that have been conditioned by heterospecifics (Bever et al. 2010). This mechanism does not require either species to be dispersal limited, because they are instead limited by patch quality (Chesson 2000b). Allowing for differences among species in dispersal ability could also increase coexistence and diversity in plant communities, for example through competition-colonization trade-offs (Chesson 2000b), in addition to the frequency-dependent coexistence that can be generated by biotic feedbacks (Bever, Westover & Antonovics 1997).

The effect of soil heterogeneity interacted with resident population turnover at the population scale (< 0.10), consistent with the qualitative predictions of theory (Schoolmaster & Snyder 2007). It is intuitive that high resident population turnover, like disturbance (Burke & Grime 1996), opens up patches for colonization, increasing invasibility. However, quantitative theory predicts that high resident turnover sometimes allows residents to track changes in patch quality, potentially leading to lower invasibility in a heterogeneous environment (Schoolmaster & Snyder 2007). We found such counterintuitive results for four of six invader species at high turnover (Table 3), suggesting that some plants may be tracking changing patch quality. For such tracking to occur, the relative quality of the patches must change over time. In other words, the environmental heterogeneity must be spatiotemporal. Our experimental design manipulated pure spatial heterogeneity by manipulating the origin of soil patches. However, some environmental drivers (e.g. those driven by plant–soil feedbacks) in this experiment might exhibit spatiotemporal environmental heterogeneity, which could influence coexistence in this system, as might be the case in other systems (e.g. Maestre et al. 2003; Allington et al. 2013). We primarily interpret resident neighbour effects as a result of competition (Fig. 3). However, for P. lanceolata invaders, residents and invaders are positively correlated in invader soil (ambient turnover treatment: Fig. S2), which is consistent with spatiotemporally variable plant–soil feedbacks mediating invader establishment. Future work will determine how much of the variance in invasibility can be explained by environmental drivers such as soil moisture and soil biota, and whether those drivers exhibit spatiotemporal variation.

This study provides the first empirical evidence that environmental heterogeneity can interact with resident population turnover to influence mutual invasibility (Schoolmaster & Snyder 2007). The population scale invasibility of these co-occurring species pairs is generally consistent with coexistence under at least some conditions. However, we do not rule out the possibility of alternative coexistence mechanisms or alternative species pairings relevant for understanding community assembly as a whole. In addition, population establishment is not equivalent to population growth rates, and future work will include demographic analysis to determine how population growth rates respond to soil heterogeneity and resident population turnover rates. We also demonstrate that population turnover can be manipulated while holding mortality/replacement constant, suggesting that perennial plant systems are particularly amenable for testing coexistence theory that depends on population dynamics. Finally, we demonstrate that plant-induced soil heterogeneity (Brandt, del Pino & Burns 2014) can influence invader establishment at the individual scale, suggesting that it could be an important driver of community assembly, potentially helping to explain apparent contradictions in the heterogeneity-diversity literature. Thus, experimental studies of the heterogeneity-diversity hypothesis should consider manipulating non-resource factors such as those driven by plant–soil feedbacks.


We thank Case Western Reserve University's Squire Valleevue and Valley Ridge Farms, including A. Locci, C. Bond and A. Alldridge, for help establishing the field experiment. G. A. del Pino, J. Hooks, L. Huffman, L. Gonzales, S. C. Leahy, A. Ubiles, C. Yu, X. Zhao and N. M. Zimmerman provided field assistance. We thank K. Abbott, M. Benard, J. Bever, C. Cope, K. Dananay, J. Koonce, K. Krynak, J. Murphy, A. Osvaldsson, R. Snyder and D. Schoolmaster for helpful conversations about these ideas. K. S. Moriuchi and two anonymous referees provided helpful comments that improved the manuscript. A.J.B. and J.H.B. were funded by start-up funds from CWRU to J.H.B. This work was also supported by National Science Foundation funding to J.H.B. (DEB 1250170). The authors have no conflict of interests to declare.

Author contribution

JHB and AJB designed and performed the experiments; AJB collected and managed the data; JHB analysed the data; JHB wrote the first draft; and AJB contributed to revisions.

Data accessibility

Invader establishment data from the 2012 field season (Burns & Brandt 2014).