Belowground biotic complexity drives aboveground dynamics: a test of the soil community feedback model



  • Feedbacks between soil communities and plants may determine abundance and diversity in plant communities by influencing fitness and competitive outcomes. We tested the core hypotheses of soil community feedback theory: plant species culture distinct soil communities that alter plant performance and the outcome of interspecific competition.
  • We applied this framework to inform the repeated dominance of Solidago canadensis in old-field communities. In glasshouse experiments, we examined the effects of soil communities on four plant species' performance in monoculture and outcomes of interspecific competition. We used terminal restriction fragment length polymorphism (TRFLP) analysis to infer differences in the soil communities associated with these plant species.
  • Soil community origin had strong effects on plant performance, changed the intensity of interspecific competition and even reversed whether plant species were limited by conspecifics or heterospecifics. These plant–soil feedbacks are strong enough to upend winners and losers in classic competition models. Plant species cultured significantly different mycorrhizal fungal and bacterial soil communities, indicating that these feedbacks are likely microbiotic in nature.
  • In old-fields and other plant communities, these soil feedbacks appear common, fundamentally alter the intensity and nature of plant competition and potentially maintain diversity while facilitating the dominance of So. canadensis.


Although soil organisms have had known dramatic impacts on agronomic plant species for centuries (de Bary, 1861) if not millennia, they have only recently been incorporated into ecological theory (e.g. Bever et al., 1997; Bever, 2003). Recent ecological studies provide compelling evidence that plant–soil feedbacks are common and may drive patterns of plant performance, relative abundance, community dynamics and exotic invasion (Klironomos, 2002; Kardol et al., 2006; Mangan et al., 2010). However, the relative importance of soil community feedbacks and their impact on key plant community interactions is largely unknown. Here we test the predictions that lie at the heart of this relatively new body of theory.

The plant feedback hypothesis: plant species ‘culture’ their own specific microbial and microzoan community within their rhizosphere which influences the performance of con- and heterospecific plants (Bever et al., 1997).

Feedbacks and competitive hierarchy hypothesis: the development of contrasting soil communities alters plant performance sufficiently to change the competitive hierarchy and thus population dynamics of coexisting species (Bever, 2003).

Local feedback interactions potentially scale up to drive patterns of relative abundance; specifically, positive feedbacks may increase plant abundance. In two landmark studies, rare species suffered more from negative feedback than common species in both temperate and tropical systems (Klironomos, 2002; Mangan et al., 2010). Plant–soil feedbacks are therefore likely to contribute to observed patterns of density dependence and species abundance (i.e. Janzen-Connell effects, Janzen, 1970; Connell, 1971; Petermann et al., 2008; Mangan et al., 2010). If these processes are indeed as pervasive and strong as the literature suggests, plant–soil feedbacks may typically modify species interactions and drive community dynamics and dominance.

Understanding the extent and strength of plant–soil community dynamics is critical to identify mechanisms that determine plant competitive outcomes, relative abundance and coexistence (Bever, 2003). Overall, little is known about belowground soil communities and their general roles in driving plant community dynamics. This is certainly understandable given the staggering diversity in belowground communities; a single cubic centimeter of soil can harbor over 6000 species of microorganisms (Torsvik et al., 2002). Although plant species can culture distinct belowground communities, for example, rhizosphere bacteria, pathogens and arbuscular mycorrhizal fungi (Westover et al., 1997; Vandenkoornhuyse et al., 2003), this is not always the case (see Burke & Chan, 2010). Currently it remains unknown how frequently plants culture distinct and species-specific belowground communities, especially in the field. More importantly, however, it is unclear the degree to which soil communities, distinct or not, benefit or harm their hosts. Nonetheless, results to date indicate that feedbacks are often negative (Bever, 1994; Reinhart et al., 2003; Kulmatiski et al., 2008) and thus plants, especially rare native species, culture communities through time that are inimical to themselves (Klironomos, 2002; Mangan et al., 2010).

Even less is known regarding the degree to which plant–soil feedbacks alter competitive relationships, yet this goes to the core of whether these feedbacks drive community dynamics. Bever (1994) found no effect of soil communities on competitive interactions, whereas other studies found compelling evidence to the contrary (Casper & Castelli, 2007; Kardol et al., 2007). Here we build upon these studies using an experimental approach that explicitly manipulates biotic soil communities and controls for abiotic effects. Perhaps most importantly, we ask whether changes in competitive interactions via plant–soil feedbacks can alter hierarchies and population projections.

We use this framework to gain insight into a long-standing pattern in ecology – the dominance of goldenrods (primarily Solidago canadensis) over dozens of other species in northeastern and mid-western US old-fields (Bazzaz, 1996; Carson & Root, 2000). Soil community feedback theory predicts that consistently abundant species may be maintained by more positive feedbacks, relative to other species. Alternatively, reciprocal negative feedback can permit coexistence in systems that would otherwise collapse to the competitive dominant (Bever, 2003). We hypothesized that more positive feedbacks, relative to other species, may help So. canadensis dominate these communities. To test these hypotheses, we used a comprehensive combination of soil community cultivation, genetic tools and a multifactorial glasshouse experiment.

Materials and Methods

We conducted these experiments at the Pymatuning Laboratory of Ecology (PLE) in northwestern Pennsylvania (Pittsburgh, PA, USA). We used four native old-field herbaceous species: Solidago canadensis L., Solidago rugosa Mill., Symphyotrichum novae-angliae (L.) G.L. Nesom and Symphyotrichum pilosum (Willd.) G.L. Nesom. The So. canadensis/altissima species complex dominates old-field communities for three or more decades after agricultural abandonment throughout large portions of the northeastern and mid-western United States and southern Canada (Bazzaz, 1996; Carson & Root, 2000; Banta et al., 2008). The other three focal species (So. rugosa, Sy. novae-angliae and Sy. pilosum) are subordinates that commonly co-occur with So. canadensis (Banta et al., 2008). We used concurrent glasshouse studies to assess whether cultured soil communities of these plant species alter plant performance and competitive ability. To support the prediction that disparate soil communities can drive plant–soil feedbacks, we subsequently used molecular approaches to quantify differences in the cultured soil communities.

Soil community cultures

Creating species-specific soil communities in the glasshouse

We cultured soil communities by growing our species from multiple, local seed sources in local field soil for 12 months in the glasshouse (following Bever, 1994). We used aseptic technique throughout the soil and seed preparation. We surface-sterilized seeds in a 5% bleach solution and then germinated these seeds in Conviron™ growth chambers in June 2004. Seedlings were transplanted into glasshouse trays filled with autoclaved silica sand and fertilized with a ½ strength Hoagland's solution twice weekly. Single seedlings (with at least two true leaves) were then transplanted into pots (20 cm × 15 cm) filled with a 3 : 1 ratio of topsoil (collected from an early successional old-field near PLE) and autoclaved silica sand. This topsoil was thoroughly chopped and homogenized before use. We used seven replicates for each of our four species, for a total of 28 pots. In the glasshouse, pots were randomly repositioned monthly and watered regularly without fertilizer. We harvested aboveground biomass following flowering in October 2004, after 4 months of growth. These individuals subsequently resprouted and flowered, and in late June 2005 we discarded aboveground and large rhizome biomass and separated soil by species. This resulting substrate is now a soil community trained by that particular plant species (e.g. Bever, 1994; Hausmann & Hawkes, 2009).

Creating species-specific soil communities in the field

Soil cultures from glasshouse studies may lack critical biotic components found only in soils grown in situ (Sykorova et al., 2007). To address this issue, we obtained soil from 5-yr old monocultures of our focal species grown in the field. We established monocultures in 2000 by planting six seedlings (from ≥ 3 local source populations) on freshly tilled Holly silt loam soils in the center of 4 × 4 m plots in a formerly agricultural old-field near PLE (Natural Resources Convservation Service). Each of the four plant species was randomly allocated to five plots for a total of 20 monocultures. Each species was allowed to clonally spread and plots were weeded to eliminate heterospecifics. On July 1, 2005, we took five to seven, 2.5 cm × 20 cm soil cores from within the monoculture of each plot. These soil samples were combined and sealed in polyethylene bags and immediately stored at 4°C until inoculum preparation. In the summer of 2008, three additional soil samples were taken from each monoculture; these samples were sent to the Agricultural Analytical Services Lab at Pennsylvania State University for soil chemistry analyses.

Preparation of inoculum pooled from field and glasshouse soil

We pooled soil from our glasshouse and monoculture experiments to create inoculum to test whether differences in soil communities could alter the performance of our focal plant species. Soil from the glasshouse and monocultures, including root fragments, was thoroughly chopped and homogenized in a 1 : 1 ratio. The soil was divided equally into two portions: one immediately stored at 4°C to use as ‘live’ soil inoculum and one autoclaved to use as sterile inoculum. To minimize abiotic effects, each ‘live’ soil inoculum was composed of equal parts of each soil community, with all but one autoclaved (Bever, 1994). For example, So. canadensis inoculum was composed of live Socanadensis soil and autoclaved So. rugosa, Sy. novae-angliae and Sy. pilosum soil communities, in a 1 : 1 : 1 : 1 ratio. A sterile control was composed of autoclaved soil communities of all four species. Additionally, we immediately stored 250 g of all live soil communities in −80°C freezer for subsequent genetic analyses.

Do soil communities alter metrics of plant performance?

We used a factorial design in the glasshouse to test if different soil communities alter the performance of So. canadensis, So. rugosa, Sy. novae-angliae and Sy. pilosum. We filled 30-cm diameter pots with a mixture of 2 kg of autoclaved local field topsoil and silica sand (3 : 1 ratio). We randomly assigned soil community and sterile control to pots. We mixed 128 g of pooled soil inoculum into the top 5 cm of soil using 1 : 16 ratio of soil community inoculum to sterile field soil. This multiple species mixture, sterilization and dilution methodology (see Preparation of inoculum above) minimizes abiotic effects and thus the impacts on plant performance are most likely due to differences in soil community. Over 6 d in early July 2005, we planted four sterile-reared seedlings (as described above) of a plant species in each pot. Our design consisted of four species grown factorially in five soil communities (So. canadensis, So. rugosa, Sy. novae-angliae, Sy. pilosum, or sterile control), replicated seven times for 140 pots. Pots were placed in a glasshouse at PLE without supplemental lighting, re-randomized every 2 wk and watered to saturation 2–3 times weekly. After 4 months, following flowering, aboveground biomass of all pots was harvested and dried at 60°C to constant mass and weighed.

Statistical analysis of soil community feedback

We tested the effects of plant species, soil community origin and their interaction on per capita aboveground biomass using analyses of covariance in PROC MIXED of SAS v9.2 (SAS Institute, 2009), with planting date as a random covariate. Separate ANOVAs were subsequently performed for each plant species with Tukey's corrected least squared multiple comparison tests. Additionally, because we wanted to examine differences between the dominant vs subordinate species response to home and away soil communities, we contrasted all subordinate species biomass (So. rugosa, Sy. novae-angliae and Sy. pilosum) together against the dominant species (So. canadensis) grown in home and So. canadensis soil and the biomass of So. canadensis in home and all subordinate soils.

We tested for positive or negative feedback with ‘home vs away’ contrast statements for paired plant species (Turkington & Harper, 1979; Bever, 1994) using the general linear models procedure in SAS. Soil community feedback (interaction coefficient, Is, sensu Bever et al., 1997) contrasts are calculated as

display math(Eqn 1)

(Aα, biomass of plant A in plant A soil; Aβ, biomass of plant A in plant B soil; Bβ, plant B biomass in plant B soil; Bα, plant B biomass in plant A soil). Values can run from positive to negative and identify the direction and magnitude of feedback. Negative feedback can arise by one or both species performing better in its competitor soil. Positive feedback is the reverse, occurring when relative performance is greater in home soil. Strong positive feedback predicts competitive exclusion while negative feedback can result in cyclically fluctuating plant performance than can maintain diversity (Bever et al., 1997; Bever, 2003). This metric is especially useful as it makes community-level predictions for the two compared species. For example, if all species responded similarly to a soil origin, feedback values (sensu Bever et al., 1997) would be near zero and therefore indicate no net feedback, while other metrics may show definitive species-level feedback.

Do soil communities change the nature of interspecific competition?

We tested whether soil communities can differentially alter the strength of interspecific competition. So. canadensis, So. rugosa, Sy. novae-angliae and Sy. pilosum were grown in pairwise competition on their own, their competitor's and sterilized soil communities. Our methodology is identical to the intraspecific experiment as already described, except two sterile-reared seedlings of each of two plant species were planted into each pot, with plant species alternated. In all, our design consisted of six pairwise plant competition combinations, three soil communities (home, competitor, sterile) and seven replicates for 126 pots. This experiment was conducted simultaneously with the intraspecific feedback experiment as already described.

Data and statistical analysis of feedback on competition

We evaluated whether plant–soil feedbacks altered the nature of intra- vs interspecific competition by using a Relative Interaction Intensity index (RII):

display math(Eqn 2)

(PC, per capita biomass of the focal plant species in interspecific competition; PM, per capita biomass of the focal plant species in intraspecific competition or monoculture (Howard & Goldberg, 2001)). The denominator is whichever value (intra- or interspecific per capita biomass) is larger. The RII metric is confined to the range −1 to + 1 and scales symmetrically c. 0. Positive values indicate that the strength of interspecific competition is greater than intraspecific competition. Negative values indicate the reverse, that is, plant growth is more limited by conspecifics, indicative of overyielding. We calculated RII for each focal plant species with each plant competitor in home, away and sterile soil communities.

We tested the effects of focal plant species, plant competitor species, soil community origin and all interactions on Relative Interaction Intensity in PROC MIXED of SAS (SAS Institute 2009), with planting date as a random covariate. Separate ANOVAs were performed at the focal plant species and focal × competitor interaction levels with Tukey's corrected least squared multiple comparison tests. We analysed feedback identically as the monoculture experiment with a factorial ANCOVA and subsequent contrast statements in SAS.

Do soil community feedbacks alter plant population projections?

Our data allows us to partially parameterize the soil community feedback model first proposed by Bever (2003) to evaluate if experimentally-derived feedback values alter classical competition predictions. The soil community feedback model modifies the Lotka–Volterra competition model as

display math(Eqn 3)

(NA, biomass (or density) of plant A; rA, population growth rate of plant A; αA, change in growth rate of plant A in home soil; βA, change in growth of plant A in competitor soil; cB, relative competitive effect of plant B on plant A; KA, carrying capacity for plant A in the absence of feedbacks and competition; and SA and SB, the relative abundances of plant A's and B's soil community, respectively). Similarly, soil community dynamics (relative change in soil community) are described by:

display math(Eqn 4)

(ν, the relative influence of plant B on the soil community).

We parameterized these models with data derived from both monoculture and interspecific competition experiments. Competition coefficients were calculated as the difference between intra- and interspecific competition biomass in sterile conditions. Feedbacks were determined from relative comparisons of biomass between sterile and specific soil communities. To isolate competition-feedback dynamics and avoid unknown parameters, all species were given identical population growth rates, carrying capacities and initial population sizes (see Fig. 4 for parameters). To test the influence of plant–soil feedbacks, we ran models with and without feedback parameters, whereby Bever's feedback model (Eqn 2) collapses to the classic population Lotka–Volterra model (Bever, 2003).

Quantifying contrasting soil inoculum cultured by our focal species

We identified differences in the bacterial and arbuscular mycorrhizal fungal communities of our original plant species inoculum by employing terminal restriction fragment length polymorphism (TRFLP) analysis (Helgason et al., 1999; Vandenkoornhuyse et al., 2003; Burke, 2008). We extracted DNA from four 0.5-g samples of homogenized soil inoculum from each plant species (mixed field and glasshouse trained soil) using a MO BIO Powersoil DNA Isolation kit (Carlsbad, CA, USA). To describe the arbuscular mycorrhizal fungal community, a c. 500-bp fragment of the small subunit rRNA sequence was amplified by PCR using primers NS31 and AM1 (Simon et al., 1992; Opik et al., 2006) following general procedures in Helgason et al. (1999). The NS31 primer was labeled with 4, 7, 2′, 4′, 5′, 7′ -hexachloro-6-carboxyfluorescein (HEX) and the AM1 primer was labeled with fluorochromes 6-carboxyfluorescein (6FAM). PCR products were cut using Hinf I and Hsp 92 (Promega) restriction enzymes. We note that NS31 and AM1 primers primarily amplify arbuscular mycorrhizal rRNA, but may additionally amplify other fungal taxa (Hausmann & Hawkes, 2009; see the Discussion section). Analyses of the bacterial communities were conducted with bacterial domain forward primer 338f and 6FAM-labeled reverse primer 926r to target the 16S rRNA gene; TRFs were generated with Mbo I restriction enzyme (Fermentas, Glen Burnie, MD, USA). These primers were used because they amplify a wide taxonomic range of bacteria and are commonly used in studies of soil bacterial communities and our protocol followed Burke et al. (2006). TRFLPs were analysed at the Cornell Bioresource Center using an Applied BioSystems (Foster City, CA, USA) 3730xl DNA Analyzer.

Statistical analysis of soil biotic communities

We generated relative TRFLP profiles for both bacterial and mycorrhizal communities, following protocols described by Vandenkoornhuyse et al. (2003). Every distinct terminal restriction fragment (TRF), identified as a peak > 1% of total peak area, was treated as identification of microbial taxa (Vandenkoornhuyse et al., 2003; Burke & Chan, 2010). Our community analysis was based on the presence vs absence of TRFs. We assessed patterns of bacterial and mycorrhizal fungi community composition with nonmetric multidimensional scaling (NMDS) ordination for soil communities cultured by each plant species (see Frank et al., 2004; Burke & Chan, 2010). We tested for significant differences among these soil biotic communities with the Jaccard dissimilarity index. We used a nonparametric, permutation-based MANOVA to test for differences in taxon composition because rare species led to violations of assumptions of normality required for parametric MANOVA (Legendre & Anderson, 1999; McArdle & Anderson, 2001). Finally, we conducted Jaccard contrasts calculated percent taxon overlap to test pairwise dissimilarity between soil communities. Analyses were conducted in R (R Development Core Team, 2011), using the vegan package (Oksanen et al., 2012) and adonis function.


Hypothesis 1: soil communities differentially alter plant performance

Soil origin caused significant and highly species-specific differences in plant performance (particularly for subordinate species) that depended upon the origin of the inoculum (plant species × soil origin interaction P = 0.035, see Supporting Information Fig. S1, Table S1). In general, Solidago canadensis performed better in away (i.e. subordinate) soil than home soil (Fig. 1a, Table S1). Conversely, the subordinate species as a group performed significantly better in their home soils than in So. canadensis soil (Fig. 1a).

Figure 1.

Monoculture growth responses and soil community feedback. Asterisks indicate significant differences or where feedback is significantly different from zero. Relevant statistics are reported in Supporting Information Table S1. Error bars indicate ± SE. (a) Biomass of Solidago canadensis and all subordinate species in So. canadensis (open bars) and subordinate species' (closed bars) soil communities. In general, all species performed better in subordinate species' soil than when grown in So. canadensis soil (So. canadensis, P = 0.0326; Subordinates, P = 0.0471). (b) Feedback of the soil community in monoculture on the growth of four plant species. Feedback is calculated as the difference in growth of plants in their own soil community and growth in each of their competitors' soil communities (IS, Bever et al., 1997). Symphyotrichum novae-angliae is abbreviated as Sy. novae.

We found significant negative feedback in monoculture for two of our six species pairs (Table S1, Fig. 1b). For one pair, (So. canadensis vs Symphyotrichum pilosum) this negative feedback occurred because each species performed better in its competitor's soil. For the other pair (Sy. novae-angliae vs Sy. pilosum) Sy. pilosum performed significantly better in its competitor's soil, while Sy. novae-angliae grew similarly in both soil origins (Fig. S1).

Hypothesis 2: feedbacks and competitive relationships

Plant–soil feedbacks were stronger in interspecific competition, relative to monoculture. Indeed, soil community origin significantly altered competitive response (as defined by Relative Interaction Intensity) for all plant species and for half of our pairwise competition trials (plant species × competitor × soil origin interaction P = 0.0059; Table S1, Fig. 2). In three cases, soil community identity changed the nature of competition, determining whether intraspecific competition was stronger than interspecific competition (Fig. 2a,c,d). Furthermore, in four cases and for all species, type and intensity of competitive interactions in live soil were significantly different from sterile controls (Fig. 2). For example, Sy. pilosum was more limited by conspecifics in its own soil community, but more limited by heterospecifics (Sy. novae-angliae) in sterile soil (Fig. 2d). We also detected significant soil community feedbacks in three out of six possible pairwise competition trials (Fig. 3b; Table S1). Solidago canadensis had increased performance in both Symphyotrichum spp. soil, resulting in greater biomass in subordinate soil (Fig. 3a). By contrast, subordinates responded similarly to So. canadensis and subordinate home soil. This pattern results in negative feedback between So. canadensis and both Symphyotrichum species (Fig. 3).

Figure 2.

Relative interaction intensity response of four plant species to competitors in conspecific, competitor, or sterile soil community. Competitor identity is labeled above the x-axis; the shaded bars indicate soil community origin (white bars, home; grey bars, competitor; black bars, sterile). For example, the leftmost three bars in (a) display the competitive response of Solidago canadensis to So. rugosa in home (So. canadensis), competitor (So. rugosa) and sterile soil communities. Capital letters indicate significant differences (< 0.05) in competitive response to soil community within plant–competitor pairs. In (d) A* indicates a Tukey's corrected P-value of 0.067. All species had a significant response to both competitor identity and soil community origin, or a significant competitor by soil community interaction. Error bars indicate ± SE.

Figure 3.

Growth responses and soil community feedback in interspecific competition. Asterisks indicate significant differences or where feedback is significantly different from zero. Relevant statistics are reported in Table S1. Error bars indicate ± SE. (a) Biomass of Solidago canadensis and all subordinate species in So. canadensis (open bars) and subordinate species' (closed bars) soil communities in competition. Only So. canadensis had increased growth in subordinate soil (So. canadensis,< 0.0001; Subordinates, P = 0.9637). (b) Soil community feedback in interspecific competition on the growth of four plant species. Symphyotrichum novae-angliae is abbreviated as Sy. novae. Sy. novae-angliae vs Sy. pilosum feedback displays a trend of significance with a P-value of 0.0683 (marked ‘T’).

The presence of soil community feedbacks reversed population projections for So. canadensis and Sy. pilosum. Classic Lotka–Volterra competition models predict Sy. pilosum to win in competition over So. canadensis (Fig. 4, solid lines). However, the feedbacks generated by these species reverse this hierarchy thereby promoting So. canadensis' rise to dominance while increasing coexistence time (Fig. 4, dashed lines). It should be noted that Sy. pilosum had only a slightly higher competition coefficient than So. canadensis in sterile conditions.

Figure 4.

Population density projections for Solidago canadensis and Symphyotrichum pilosum parameterized with experimentally derived competition and soil community feedback coefficients. Projections are shown with (dashed lines) and without (solid) feedbacks. In all simulations, K = 100, r = 0.5, CSC = 2.4 and CSP = 2.8. Species and soil communities were begun with equal frequencies. Sy. pilosum had a slightly greater relative impact on So. canadensis biomass and thus is predicted to win in competition. However, negative feedback (αSCSC = −0.097, αSCSP = −0.04, βSPSP = −0.096, βSPSC = 0.192, ν = 1) reverses this hierarchy and increases coexistence duration.

Plant species culture disparate soil communities

TRFLP analysis revealed a total of 39 terminal restriction fragments (TRFs) of arbuscular mycorrhizal fungi using restriction enzyme Hsp 92 and the AM1 primer and 41 TRFs using Hinf I and the NS31 primer in the rhizosphere of the four plant species. We also found a total of 31 bacterial TRFs using restriction enzyme Mbo I in the rhizosphere of the four plant species. For simplicity, we present data for the arbuscular mycorrhizal fungi community using the AM1 primer; nonparametric MANOVA and ordination analyses results from the NS31 primer were qualitatively identical. Nonparametric MANOVA and ordination analyses demonstrated that each plant species cultured highly distinct soil bacterial (F3,15 = 4.25495, < 0.001) and mycorrhizal communities (F3,15 = 4.03485, < 0.001, Fig. 5, Table S2). Pairwise Jaccard contrasts and nonparametric multivariate analyses confirmed that these communities were statistically distinct (Table S2). Of 12 pairwise comparisons, only the mycorrhizal communities of Sy. novae-angliae and Sy. pilosum were not significantly different (P = 0.097). Indeed, soil origin explained over 50% of the variation in both mycorrhizal and bacteria communities (R2 = 0.5022, 0.5154, respectively). Taxon overlap averaged 38% and 61% across plant species pairs for bacterial and fungal communities, respectively (see Table S2). Overall soil community TRFLP characteristics and taxa relative abundance data are presented in Tables S3 and S4, respectively. By contrast, soil chemistry was relatively consistent across plant species. Although So. canadensis had higher concentrations of soil potassium than the other focal species (P = 0.004), total% nitrogen, phosphorus, magnesium, calcium and phosphate concentrations, as well as pH, were statistically similar across species (see Table S5).

Figure 5.

Nonmetric multidimensional scaling ordination of bacterial (a) and arbuscular mycorrhizal fungi (b) communities from Solidago canadensis, So. rugosa, Symphyotrichum novae-angliae and Sy. pilosum using terminal restriction fragment length polymorphism (TRFLP) analysis. Each numbered datum represents a single soil sample; envelopes are drawn around replicates of soil communities of each plant species. Soil origin created consistently different belowground communities (see Table S2).


Overall, we found compelling evidence for the fundamental tenets of soil community feedback theory. Soil community feedbacks were strong enough to influence plant performance in monoculture, and alter the intensity of interspecific competition and reverse population projections over time. Additionally, we draw strong inference that plants, even closely related species from the same genus and successional stage, culture disparate soil microbiotic communities that can drive feedback. We also found compelling evidence that negative feedbacks may reinforce the dominance of a competitively superior species (Klironomos, 2002) and fundamentally alter population projections. Finally, and perhaps most importantly, our results demonstrate unequivocally that feedbacks can change the nature of competitive interactions causing intraspecific interactions to be stronger than interspecific interactions (and vice versa), depending upon the origin of the soil community. Such reversals in limitation are fundamental to species coexistence theory (Chesson, 2000).

Soil communities differentially alter plant performance

The performance of every plant species was dependent on soil community origin (Figs 2a, S1; Table S1), suggesting that species-specific, plant–soil interactions are common in these plant communities. Moreover, Solidago canadensis performed relatively better when grown in the soil of competitors (Fig. 1a) with which it commonly co-occurs, suggesting these feedbacks could facilitate invasion into competitor territory by this dominant species. We found significant negative feedback in two pairs of species in monoculture (Fig. 1b), caused by superior growth in the competitor's soil community. Overall, our results add to a recent and growing literature verifying that soil communities commonly alter plant performance and often result in negative feedback (Bever, 1994; Casper & Castelli, 2007; Kardol et al., 2007; Casper et al., 2008; Kulmatiski et al., 2008).

Changes in plant performance are likely an explicit response to differences between soil communities. That said, identifying the mechanism(s) of feedback using this methodology is problematic because our black box approach collapses all belowground interactions into plant performance. We simply do not know the causal or interactive agents of the soil community that influence performance and competitive relationships. However, we believe that differences in soil microbiotic communities are driving these feedback patterns. By inoculating standardized, sterile soil with relatively small amounts of soil cultured individually by all species and sterilizing all but one, our protocol minimizes the possibility of confounding soil microbial interactions with other feedback processes (e.g. resource availability, chemical profile, aggregation, see Bever, 1994; Ehrenfeld et al., 2005; Kardol et al., 2006; Casper & Castelli, 2007). Additionally, soil inoculum sources were remarkably similar in soil chemistry and feedback responses inconsistent with abiotic differences. Although it is possible that autoclaved inoculum had altered soil structure and chemical profiles to influence plant performance and biomass allocation, the 15 : 16 ratio of sterile bulk soil to inoculum is expected to strongly dilute these effects. Thus, the altered plant performance across observed soil origins is most likely due to differential responses to soil microbiotic communities. The negative feedbacks observed here may be linked to an accumulation of pathogenic bacteria or fungi, poor nutrient cyclers, or host-specific inefficient mutualists, such as arbuscular mycorrhizal fungi (Bever, 2002; Packer & Clay, 2004). Additionally, indirect negative feedbacks are also possible with mismatches of host-specificity and benefit with mutualists or detriment with pathogens (Bever et al., 1997).

Extensive soil community feedbacks are magnified under competition

Feedbacks were more common and, in general, stronger in interspecific competition than monoculture. Indeed, in a post-hoc test, we found relative biomass deviation between home and away soil increased over 30% in competition compared with monoculture (F = 7.12, < 0.0084). Overall, half of all competitive interactions were significantly altered by soil community identity and three out of six species pairs exhibited significant feedbacks. As in monoculture, the dominant species, So. canadensis performed better in the soil cultured by two of the subordinates (Symphyotrichum spp.), relative to home soil. Bever (2002) and van der Putten et al. (1997) have demonstrated how an accumulation of pathogens or mycorrhizal fungi can lead to negative feedback and should diminish the performance of the abundant species. However, negative feedbacks were never strong enough in our system to alter the competitive hierarchy between the dominant species (So. canadensis) and the subordinates (Fig. 2b,c,d). Specifically, subordinates responded to So. canadensis competition similarly in all soil communities. Thus, subordinates may not be able to invade (or reinvade) the dominant's territory, whereas the performance of So. canadensis is enhanced in the subordinate's soil. By contrast, this was not the case with another large-statured goldenrod species (So. rugosa) that can sometimes form dense, but small and spatially patchy, stands in old-fields. Here we found a significant positive feedback between So. rugosa and Sy. pilosum that predicts the extinction of So. rugosa largely because of Sy. pilosum's greater performance in home soil in competition. Concordantly, Sy. pilosum has consistently higher population densities than So. rugosa in old-fields (Banta et al., 2008), with feedbacks possibly providing a mechanism for So. rugosa's low abundance and patchy distribution.

Overall, one unequivocal take-home message is that soil identity determined whether intraspecific interactions were stronger than interspecific interactions. In fully half of comparisons, soil community identity significantly altered competitive intensity and in three instances completely reversed the strength of intra vs interspecific competition (Fig. 2). For example, the strength of interspecific competition decreased drastically for both So. rugosa and Sy. pilosum in the soil community of Sy. pilosum, likely to result in longer coexistence (Fig. 2b,c). While the few studies of feedback effects on competition indicate per-capita differences between monoculture and competition responses (e.g. Bever, 1994; van der Putten & Peters, 1997; Casper & Castelli, 2007; Kardol et al., 2007; Petermann et al., 2008; but see Callaway et al., 2004), to our knowledge, ours is the first to demonstrate that plant–soil feedbacks can be so strong that they determine whether competition is stronger within or between species. This is indeed a cornerstone mechanism of species coexistence: when species are more limited by themselves than by heterospecifics (Chesson, 2000). For example, the influence of plants on soil communities may create shifting, fine-scale mosaics of performance, where the strength of competitive interactions shift between intra- and interspecific. This generates spatial and temporal heterogeneity, even within a seemingly homogeneous environment and may serve to favor one species, then another, et cetera over the scale of a few meters. If soil community feedbacks are strong or competitive abilities are similar or both, traditional competitive hierarchies may continuously reverse depending on neighbor and soil legacy, potentially allowing stable coexistence for numerous species (Chesson, 2000; Petermann et al., 2008).

Congeneric species support disparate soil communities

There is growing consensus that different plant species culture disparate soil communities, from pathogens and rhizosphere bacteria to mycorrhizal fungi (Grayston et al., 1998; Holah & Alexander, 1999; Vandenkoornhuyse et al., 2003; Batten et al., 2008). Our findings reaffirm this pattern for bacterial and mycorrhizal fungi. There are likely complex abiotic, biotic, host-specific and trophic interactions that influence soil community constituents and diversity, and recent research indicates that even genotypic differences can alter root endophytic bacterial communities in Arabidopsis thaliana (Bulgarelli et al., 2012; Lundberg et al., 2012).

Similarly, our congeneric species, even from the same community and successional stage, supported significantly different belowground communities, with only one exception (Fig. 5; Table S2). Our results indicate that plants species create fine-scale community variation across very small spatial scales. Indeed, many plant communities, particularly old-fields, are characterized by distinct patches of long-lived herbaceous plants (Carson & Pickett, 1990; Bazzaz, 1996), potentially facilitating the formation of zones of distinct soil communities. This adds a definitive layer of complexity to our view of communities. Specifically, as plant species, even at the individual level, differentially deplete soil resources (Tilman & Wedin, 1991), they also culture novel and complex biotic communities belowground.

We acknowledge that the molecular methods we used here to describe the soil microbial communities are only implicative. By subsampling homogenized soil inoculum, we certainly compiled variation within plant species. Additionally, the primers we used extend an imperfect view of bacterial and mycorrhizal diversity. The ‘AMF’ primers we used may additionally amplify a minority of other fungal taxa (Burke, 2008; Hausmann & Hawkes, 2009) and thus community dissimilarity may be due to differences in arbuscular mycorrhizal fungi, ascomycetes or a select few other fungi, or a combination of these groups. It is also very likely that the soil bacterial and fungal taxa we amplified with our primer sets and our analysis are only a subset of total microbial diversity. We simply use these fungal and bacterial TRFLPs as indicators of the larger soil community and did not attempt to link feedback to specific soil taxa. Indeed, we do not know if feedback is driven by the accumulation of specific pathogen(s) (e.g. Reinhart & Callaway, 2006) or via complex trophic interactions. However, we draw strong inference that the consistent soil community dissimilarity observed across our four plant species is indicative of plants culturing disparate microbiotic communities in the field, and these communities can drive belowground feedback.

Feedbacks, dominance and old-field communities

Old-field communities have seemingly conflicting characteristics: dominance by a few species in an otherwise highly diverse community over small spatial scales (Carson & Pickett, 1990; Bazzaz, 1996). Solidago canadensis may be the most abundant perennial forb in the northeastern United States and southeastern Canada (Carson & Root, 2000; Banta et al., 2008 and citations therein) and its repeated pattern of dominance has defied explanation, although it is possibly linked to its ability to cast deep shade (Werner et al., 1980; Carson & Pickett, 1990; Carson & Root, 2000; Banta et al., 2008). Here, soil community feedback may provide mechanisms that facilitate goldenrod invasion as well as potentially maintaining diversity within the system by decreasing competitive intensity. Solidago typically invades 2–5 yr after agricultural abandonment (Bazzaz, 1996; Carson & Root, 2000). The relatively superior growth of So. canadensis in both Symphyotrichum soils likely hastens its invasion into early successional aster and grass dominated old-fields. The relatively poor performance of So. canadensis in home soil (Figs 2a, 3a, 4a) is predicted to decrease its competitive ability over time and thus its suppression of heterospecifics. Indeed, the replacement of Jacobaea vulgaris over time in European old-field is likely driven by negative feedback (van de Voorde et al., 2011). However, if negative feedback decreases competitive intensity without reversing the competitive hierarchy, it may allow the persistence of competitive subordinate species (Bever, 2003) without altering the dominance of So. canadensis. We acknowledge that this mechanism contrasts with our prediction that So. canadensis should have generally positive feedbacks, but it suggests how So. canadensis can dominate old-fields and yet coexist with many plant species.

This study supported the predictions of soil community feedback theory (Bever, 2003), by demonstrating that species-specific soil community feedbacks reversed competitive hierarchies and increased the duration of species coexistence (Fig. 4). This is a simplified illustration of how these species interact, but provides an empirically-generated demonstration of the potentially critical role of feedbacks in natural communities. Our plant–soil feedback population projections closely mirror natural species abundance over time (across > two dozen fields): So. canadensis increases in abundance to dominance over time, while Sy. pilosum is eventually excluded by year 30 (Banta et al., 2008). By contrast, the only other plant–soil population projections found rapid plant extinction when pathogen feedbacks were considered, helping to explain the invasion by Ammophila arenaria into Pacific dune communities in the United States (Eppinga et al., 2006). These examples highlight the potentially dramatic role plant–soil feedbacks may play in community assembly and population dynamics. These feedbacks need not be strong to alter community dynamics – subtle changes to competitive hierarchies or access to soil resources may result in striking changes in population persistence and community structure – especially if there are minor differences in competitive ability and character displacement. Indeed, with dozens of coexisting species altering soil communities, plant–soil feedbacks may create temporal and spatial heterogeneity that fundamentally change local competitive outcomes and promote coexistence (Ricklefs, 1977; Reynolds et al., 2003).


Soil community feedbacks are common, likely microbiotic in nature, and strong enough to alter performance and competition in these old-field species, and across communities and habitats. Although negative feedbacks generally promote species replacement (Kardol et al., 2007), they also alter competitive interactions and facilitate invasion into competitors' territory. In our system, plant–soil feedbacks shape plant community dynamics and possibly promote coexistence, in our case, even in spite of a competitive dominant.


We are grateful to James D. Bever, Scott J. Meiners, Jill T. Anderson and Henry B. Schumacher for discussion and statistical advice, as well as Scott A. Mangan, Wim H. van der Putten, the Charles E. Mitchell Lab and anonymous reviewers for their insightful comments. We thank Grace S. Lloyd, Siarhei Tsymbalau and Garland P. Waleko for their field and glasshouse assistance, and Roxanne Fisher of Chatham University for glasshouse use. This work was funded by the G. Murray McKinley Research Fund of the Pittsburgh Foundation, a NSF Doctoral Dissertation Improvement Grant (#0508012) and the Corning Institute for Education and Research. This is contribution number 281 to the Pymatuning Laboratory of Ecology.