Plant community evenness responds to spatial plant–soil feedback heterogeneity primarily through the diversity of soil conditioning

conditioned soils, heterospecifics typically outperformed the focal species. In addition, there was a trend for increasing community evenness from uniform, via fine- grained to coarse-grained mixed- conditioned soils, but this was not significant. community Our data demonstrate that PSFs play a role in plant evenness. Across mono- conditioned soils, PSF to altered competitive hierarchies. on soils conditioned by multiple species, competitive ability among species was more similar and this led to higher plant evenness. The spatial distribution of the heterogeneity, on the other hand, did not significantly affect plant evenness. Our data therefore show that community evenness was more strongly related to the number of plant species that conditioned the soil than the spatial distribution of the PSF heterogeneity. Future studies need to investigate the importance of PSFs in the field across plant life stages and multiple generations.


| INTRODUCTION
A long-standing question in ecology is how high plant diversity is maintained at local spatial scales (Hutchinson, 1961;Wilson, Peet, Dengler, & Pärtel, 2012), as competitors are known to exclude one another (Hardin, 1960). Differences in competitive abilities lead to a decline in community evenness and eventually to local extinction of subordinate species (Booth & Grime, 2003;Silvertown, 2004;Wilsey & Polley, 2004). Several authors have proposed that at small spatial scales, plant antagonists, particularly soil-borne pathogens, may act as key drivers of plant evenness and diversity (Bennett & Cahill, 2016;Bever, Mangan, & Alexander, 2015;Bradley, Gilbert, & Martiny, 2008;De Kroon et al., 2012).
When growing in soil, plants induce changes in the composition of the soil community (Bezemer et al., 2010;Lundberg et al., 2012) and these changes, in turn, affect plant performance. This phenomenon is known as plant-soil feedback (PSF; Bever, 1994Bever, , 2003Van der Putten et al., 2013). Most direct PSF effects, that is the effect of growing on self-conditioned soil, are reported to be negative, and this may prevent species from becoming mono-dominant in the community (Kulmatiski, Beard, Stevens, & Cobbold, 2008;Petermann, Fergus, Turnbull, & Schmid, 2008). The particular soil organisms that cause the PSF effect may vary between soils and may be different for each plant species in the community . Furthermore, interactions with other soil organisms can strongly alter the effects of mutualists and pathogens on plants (Bradley et al., 2008;Hersh, Vilgalys, & Clark, 2012;Morris et al., 2007). PSF studies quantify the net effect of changes in the whole soil community on plant performance and as such incorporate the complex web of interactions below-ground (Bever, Westover, & Antonovics, 1997; Van der Putten et al., 2013).
Theoretical models of spatial plant communities show that PSF effects can mediate plant diversity if they are highly localized in space (Abbott et al., 2015;Bonanomi, Giannino, & Mazzoleni, 2005;Fukami & Nakajima, 2013;Mack & Bever, 2014). Recent empirical work has confirmed that spatial heterogeneity in PSFs can affect plant performance (Brandt, De Kroon, Reynolds, & Burns, 2013;Burns, Brandt, Murphy, Kaczowka, & Burke, 2017;Hendriks, Visser, et al., 2015;Wubs & Bezemer, 2016), seedling establishment (Burns & Brandt, 2014) and can alter competitive hierarchies (Hendriks, Ravenek, et al., 2015). Furthermore, individual grassland plant species growing in open field communities are known to have specific soil communities that are distinct from neighbouring individuals of other species (shown using 5-cm-diameter cores; Bezemer et al., 2010). However, thus far PSF studies have been limited to inferences based on data from monocultures or pairs of plant species. Empirical evidence of the impact of heterogeneity of plant-soil feedback on plant diversity in F I G U R E 1 Experimental design. Two plant communities (mix 1 and mix 2) were planted each on seven conditioned soils with different levels of conditioning diversity and at different spatial grain. Soils were either homogeneous (uniform) or heterogeneous (fine-or coarsegrained heterogeneity). Heterogeneous soils were conditioned by the same four species as in the respective plant mixture. The homogeneous soils (uniform) were either conditioned by one species (mono-conditioned) or simultaneously conditioned by four species in mixture (mixedconditioned). Ac = Agrostis capillaris; Fr = Festuca rubra; Hr = Hypochaeris radicata; Jv = Jacobaea vulgaris; Lc = Lotus corniculatus; Tp = Trifolium pratense communities consisting of more than two species is lacking (Hendriks, Ravenek, et al., 2015). In addition, spatial PSF heterogeneity consists of two components that have hitherto not been teased apart: conditioning diversity (CD) and spatial grain ( Figure 1).  (Booth & Grime, 2003;Hillebrand, Bennett, & Cadotte, 2008;Wilsey & Polley, 2004). However, rates of local extinctions have been shown to be higher in communities with lower evenness, and differences in evenness are thus expected to translate into differences in species richness over time (Wilsey & Polley, 2004).
In this experiment, we explicitly tested whether the effects of CD or spatial grain were the dominant drivers of PSF effects on community evenness.
We predicted that plant evenness would be higher in mixedconditioned treatments. This is because in uniform soil conditioned by one species (mono-conditioned), heterospecific plants are expected to become dominant as they would not encounter their own soil-borne antagonists within their rooting zones (Casper, Schenk, & Jackson, 2003). In contrast, in soils conditioned by multiple species, each species would be kept in check by its own set of soil-borne antagonists.
Therefore, we predicted that the competitive abilities among species in a community would be more equal in mixed-conditioned soils compared to mono-conditioned soils and that this led to a higher community evenness.
In addition, we manipulated the grain of the spatial heterogeneity while keeping the ratios of the differently conditioned soils the same (i.e. all mixed-conditioned soils). Some plant species are known to be competitively superior when abiotic soil resources are distributed heterogeneously, because of their superior root placement ability (Gazol et al., 2013). Likewise, specific root placement ability in response to heterogeneous PSFs also differs among species (Hendriks, Ravenek, et al., 2015). This suggests that in heterogeneous mixed-conditioned soils, some species may have a competitive advantage as they can place their root preferentially in patches with the least negative PSF, suggesting that evenness will decline in heterogeneous conditions (evenness: coarse-grained < fine-grained < uniform). However, in contrast to soil-borne resources, soil biota are mobile to some extent and this may blur the initial heterogeneity (e.g. after a plant, or a root, dies).
If the latter effect is dominant, then each species is equally likely to encounter its enemies regardless of their exact spatial location (Wubs & Bezemer, 2016). In that case, we predict that spatial grain would have no effect on plant performance and evenness, and therefore, we predicted that uniform soils conditioned by multiple species (mixedconditioned) would have the same effect on plant evenness as spatially heterogeneous treatments.

| MATERIALS AND METHODS
To test our hypotheses, we conducted a glasshouse PSF experiment where we grew two plant communities on four soil conditioning treatments, where we explicitly manipulated the spatial PSF heterogeneity ( Figure 1). This study builds on the work reported in Wubs and Bezemer (2016)  day:night, 50%-70% relative humidity) for 8 weeks. Subsequently, shoot biomass was clipped and all root systems were removed from the soil of each pot. Soil from containers in which the same species had grown and that were a priori allocated to the same soil replicate were pooled and carefully homogenized. To obtain a sufficient amount of soil for the experiment, each of the soil replicates was mixed with sterilized (>25K Gray gamma radiation; Isotron, Ede, the Netherlands) field soil collected from the same site in an 8.4:1.6 (conditioned: sterile w:w) ratio. From each of the homogenized soil replicates, a sample (200 g) was taken for chemical analysis after addition of the sterilized soil. We measured mineral nitrogen (KCl extraction), PO 4 (P-Olsen extraction) and soil organic matter (ashed at 430°C for 24 hr) content as well as soil acidity (in 1:2.5 w:w dry soil:water suspensions; see Table   S1). Furthermore, three soil samples, one per soil replicate, were taken from each of the mono-conditioned soils (n = 18) and analysed for differences in fungal community composition using terminal restriction fragment length polymorphism (T-RFLP) analysis of the ITS marker (see Methods S1 for the protocol).

| Phase 2: Feedback phase
In the test phase, three different levels of spatial grain were created (spatially homogeneous, spatially heterogeneous coarse-grained and spatially heterogeneous fine-grained; Figure 1). Each container (length × width × height: 26 × 22 × 22 cm) was divided using a custom-made metal grid into 4 × 4 cells, each with a surface area of c. In each container, independent of the treatment, all 16 grid cells were filled individually and any given grid cell was always filled with a single conditioned soil type. Each container was filled with 2.5 kg sterilized gravel (quartz, 4-8 mm) and then with 8 kg of conditioned soil (500 g per grid cell). All containers were filled with conditioned soil in the same way: weighing 500 g of the appropriate conditioned soil type and then carefully pouring the soil into the respective grid cell.
Immediately after filling the containers, the metal grid was removed so that during the test phase, the soil patches in each container were in full contact.
For the homogeneous treatment, all cells in a container were filled with one conditioned soil (either mono-or mixed-conditioned), while for spatially heterogeneous treatments (coarse-and finegrained), grid cells were filled with soil mono-conditioned by four different species (Figure 1). The four soils in the fine-grained treatment were applied following a Latin square design, while for the coarse-grained treatment, four contiguous square blocks of four cells each were created in each container. The two spatially heterogeneous treatments (coarse-and fine-grained) were created with two different mixes of soil conditioned by four plant species (soil mix). Soil mix 1 consisted of soils conditioned by A. capillaris, H. radicata, J. vulgaris, and L. corniculatus; soil mix 2 consisted of A.
capillaris, F. rubra, J. vulgaris and T. pratense. Consequently, both soil mixes had at least one representative each of the grass, forb and legume plant functional types. Containers in the mixed-conditioned uniform treatment received the soil that was simultaneously conditioned by four species in all grid cells. As the mixed-conditioned uniform soils were homogenized at the end of the conditioning phase, as was true for all conditioned soils, we expect that any spatial differences will have been evened out and therefore consider it as a spatially uniform treatment. Conditioned soils from all six focal species were used separately to create spatially homogeneous containers with mono-conditioned soil.
After preparation of the soil treatments, the conditioned soils were planted with two mixtures of four plant species, which were the same as the species that conditioned the two soil mixes mentioned above.
Data from both mixes were analysed only when growing on soils in their own mix. Consequently, each plant mixture grew on seven soilby-spatial heterogeneity treatments (four mono-conditioned uniform, one mixed-conditioned uniform, one coarse heterogeneous and one fine heterogeneous). The four species were planted in a Latin square design, which was selected randomly with the constraint that each plant species would be planted on all four soils in the spatially heterogeneous treatments (i.e. a Graeco-Latin square for the fine-grained heterogeneous treatment). The whole set-up was replicated three times, using the three independent soil replicates. In total, there were 42 containers in the test phase (7 soil treatments × 2 plant communities × 3 replicates).
Each container was planted with 32 seedlings, planting two individuals of the same species into each grid cell (each seedling 1 cm from the grid cell mid-point). The experimental design ensured that all plant species were grown on all soils in the heterogeneous treatments.
Seeds were germinated in the same way as in the conditioning phase.
Seedlings that died upon transplantation were replaced once during the first week. The containers were placed in the glasshouse in a complete randomized design under the same conditions as during the conditioning phase and allowed to grow for 8 weeks. The soil was kept moist by regular watering (two or three times per week depending on evapotranspiration rates). After 8 weeks of growth, above-ground plant biomass was clipped flush with the soil, dried (72°C, 48 hr) and weighed separately per grid cell for each of the containers (i.e. 16 observations per container, with known locations of each observation within the container).
Below-ground biomass was sampled by inserting a soil corer (Ø 3.3 cm) into the middle of a grid cell and gently pushing it to the bottom of the container. While extracting the corer, it was made sure that all soil in the column, down to the gravel underneath, was collected. To make sure the soil cores were taken exactly in the middle of each grid cell, a metal grid (the same dimensions as before, but only 1 cm high) was placed on top of the soil when taking soil cores.
Roots were extracted from the soil cores by careful washing over a sieve (2 mm mesh) and subsequently dried and weighed. For the spatially heterogeneous treatments (coarse-and fine-grained), all 16 grid cells were sampled, while eight cores were taken from the spatially homogeneous treatment. Roots could not be identified to species level and so root biomass values were summed per container to estimate total root biomass.

| Data analysis
We calculated the evenness index (J′, this is Shannon diversity rescaled to 0-1 by dividing by the natural logarithm of the number of species in the sample) based on the shoot biomass of each species present in each container as a measure of plant diversity (Pielou, 1966). Both plant evenness and total shoot and root plant biomass in each container were analysed with simple fixed-effects models. In each case, we analysed two models, one comparing the effect of CD (only including the mono-and mixed-conditioned uniform treatments) and one directly comparing the effect of spatial grain when soils were mixed-conditioned. In both analyses, plant mixture and the interaction between plant mixture and, respectively, CD or spatial grain were included as fixed effects. When the spatial grain effect was significant, we used planned contrasts to compare fine-and coarse-grained heterogeneity to the mixed-conditioned uniform treatment.
Shifts in competitive hierarchies were analysed using relative abundance of each species per container. Relative abundance was calculated as the ratio of shoot biomass of a given species to the total container shoot biomass. These data were analysed per community using linear mixed models (LMMs) with container as a random factor.
Test plant species, conditioned soil and level of spatial heterogeneity and their interactions were included as fixed factors.
To examine how PSF effects change with spatial heterogeneity, we analysed differences in shoot biomass as well as relative abundance (shoot biomass per grid cell relative to total shoot biomass in the container) on grid cell level using LMMs. The uniform mixed-conditioned soil treatment was excluded from these analyses because in this treatment the soil effects on plant performance could not be attributed to individual species that had conditioned the soil. These models included random effects for container and grid cell. The grid cell factor was introduced to account for positional effect within containers, but given the rotational symmetry in the within container design, the grid cell factor had three levels ( We analysed the effects of PSFs on plant competition in two ways. First, we analysed whether PSF effects were strong enough to alter competitive hierarchies. To do so, we calculated the number of rank reversals within both plant mix 1 and 2 across all six possible pairs of mono-conditioned uniform soils. We ranked the species based on their relative abundance within each container and calculated the number of reversed ranks among all pairs of conditioned soils, always directly comparing experimental units from the same soil replicate. We summed the number of rank reversals across all pairs and replicates and used the χ 2 distribution to test whether the number of rank reversals was more or less than 50% (Kitajima & Bolker, 2003). We interpreted significantly more rank reversals than 50% as evidence that PSFs altered the competitive ranking of the plants species.
In the second analysis, we assessed how the differences in species four values per container). Subsequently, the absolute differences between the highest and lowest CA across the species in each container were taken as a measure of the spread in CA within each community. These data were analysed in the same way as plant evenness (see above).
In some of the grid cells, seedlings died in the course of the experiment even after the first week. Seedling mortality can be an integral part of plant responses to PSF. We analysed seedling mortality at grid cell level, where we treated mortality as a binary variable, which takes the value of 1 when one or both of the plants of a grid cell had died. Plant mortality data were analysed using generalized linear mixed models (GLMMs) with a binomial error distribution. These data were analysed per plant mix with the same random-and fixed-effects structure as the LMMs described before.
Differences in abiotic soil conditions and shoot biomass at the end of the conditioning phase were tested with one-way ANOVAs.
Spearman correlations were used to assess the relationship of the soil abiotic variables at the end of the conditioning phase and the shoot biomass at the end of the test phase. These correlations were calculated for each plant community separately using all uniform soils that corresponded with the plant community (n = 15).
Differences between soils in fungal community composition (T-RFLP data) were visualized using non-metric multidimensional scaling (NMDS) and tested using a multiple-response permutation procedure. Prior to analysis, we removed singleton loci from the T-RFLP data. To test whether plant and fungal community composition were correlated, we calculated a community dissimilarity matrix (Bray-Curtis index) for both the plant and the T-RFLP data and tested their association using a Mantel test. Community dissimilarity was calculated for the mono-conditioned uniform treatments, and we only compared experimental units that occurred within the same plant mix (i.e. 4 mono-conditioned uniform soils × 3 replicates per plant mix). We pooled these values into a single Mantel analysis, where permutations were restricted within plant mix (999 permutations; Spearman's rho was used as the test statistic). The same analysis was performed for plant community composition and the dissimilarities in abiotic conditions (Euclidian distance).

| RESULTS
Plant evenness was higher in spatially uniform containers with soils that were conditioned by more species (mixed-conditioned) than in uniform containers with soil conditioned by a single species  Table S2). Evenness in the mixed-conditioned soils was F I G U R E 2 Plant evenness as a function of spatial PSF heterogeneity (a, b) and within-community differences in competitive ability (c, d). The top panels show plant evenness (M ± SE) in the four spatial PSF heterogeneity treatments for both plant mixtures (a, b). Soils were either spatially homogeneous (uniform) or heterogeneous (fine-or coarse-grained) and conditioned by either one (mono) or four (mixed, fine and coarse) species. Results of statistical analyses are indicated (cf. Table S2). The bottom panels show the relationship among plant evenness in both plant mixtures as a function of the withincommunity differences in competitive ability measured either relative to plant performance in monocultures (c, CA1) or relative to a perfectly even community (d, CA2). In both cases, there was a significant negative correlation (Spearman's rho = −0.84, p < .0001, and Spearman's rho = −0.93, p < .0001, for (c) and (d) (Table S2). These data suggest that the diversity of conditioning (mono-vs. mixed) has a larger impact on plant evenness than spatial grain (uniform, fine-or coarse-grained) itself.
On mono-conditioned uniform soils, heterospecific species often outperformed the plant species which self-conditioned the soil in terms of relative biomass production (Figure 3). This led to altered competitive hierarchies, indicated by significantly more rank reversals among species in the communities than expected by chance (47 of 72 potential rank reversals took place in both mix 1 and 2; 2 1 = 6.72, p = .01 in both cases; Table S3). Within the same plant community, different plant species gained dominance in soils that were conditioned by different monocultures (Figure 3). However, H. radicata (mix 1) and F. rubra (mix 2) were exceptions to the general pattern, as they were still competitively superior on their own self-conditioned soil ( Figure 3). Importantly, however, the performance of all species in heterogeneous soils was always intermediate to the best and the worst performance in the uniform mono-conditioned soils ( Figure 3). Furthermore, the differences in CA between the species within a community were smaller in mixed-conditioned (uniform, fine-and coarsegrained) than in mono-conditioned uniform soils (Figures S1 and S2;   Table S4), and a larger difference in CA led to a lower community evenness (Figure 2c,d).
In uniform mono-conditioned soils, five out of six species experienced significant negative PSF in the mixed plant communities based on shoot biomass (Table S5; four of six based on relative abundance; cf. Table 1). Grasses performed worst on grass-conditioned soil and better on dicot-conditioned soil (Figure 4a,b; Table 1). Dicots typically showed the reverse pattern, although they also performed well on unrelated dicot-conditioned soils (e.g. J. vulgaris grown in L. corniculatus soil performed better than on H. radicata soil; Figure 4a).
In the spatially heterogeneous soils, these patterns were altered and grasses sometimes had the highest biomass on grass-conditioned or even self-conditioned soils (e.g. F. rubra on A. capillaris soil in Coarse and F. rubra soil in Fine; Figure 4d,f). Similarly, forbs did not necessarily perform best on grass-conditioned soils (Figure 4) in the heterogeneous treatments. In general, direct PSFs, quantified as F I G U R E 3 Competitive hierarchies across the different conditioned soils for two plant mixtures. Relative abundance (M ± SE; shoot biomass) of the four plant species per treatment is shown, and the species are ranked from high to low abundance per treatment (a, b). The relative abundances sum to 1 per soil treatment (i.e. all test species in a given soil treatment). Different letters indicate significant differences among bars tested per soil treatment (see Table S9 for overall analyses). Hatched bars indicate the plant species in the mixture that grew on self-conditioned soil. The grey shading in the mixed-conditioned soil treatments (the right three sets of bars) indicates the highest-to-lowest performance range for each focal species on the four uniform mono-conditioned soils (the left four sets of bars). In all cases, the relative abundance of the focal species was not significantly different from this range. For abbreviations, see Figure 1  own-foreign contrasts, became less strong and non-significant in spatially heterogeneous soils (Figure 4; Table 1b and Table S5). Only F. rubra (mix 2) and J. vulgaris (mix 1) in Coarse showed significant negative direct PSF in heterogeneous soils (Table 1b). We conducted these analyses on both the relative and absolute shoot biomass, and this led to qualitatively the same conclusions (cf. Figure 4 and Figure   S3; Table 1 and Table S5).
Plant mortality was low in general, but varied among the plant species ( Figure S4; Table S6). In mix 1, seedling mortality of different species responded to spatial heterogeneity, with H. radicata having higher mortality in heterogeneous conditions, while the other species generally had lower mortality in heterogeneous soils. Mortality of J. vulgaris in mix 2 was elevated substantially on self-conditioned soils, which reflects its known strong negative direct PSF.
Both total community shoot and root biomass in the mixed plant communities were not affected by the spatial heterogeneity treatments ( Table S7a,b). Soil conditioning altered soil nitrogen and acidity (Table S1), but community biomass in the test phase was not related to the abiotic soil variables or to shoot biomass in the conditioning phase (Table S8). Soil conditioning did lead to clear differences in fungal community composition among the six plant species; conditioning by grasses, in particular, led to fungal communities that were different from the communities created by forbs or legumes ( Figure S5; permutation F = 1.57, p = .002). Moreover, plant and fungal community composition were significantly correlated across mono-conditioned uniform treatments (Mantel r = 0.189, p = .024; Figure S6a-c), while this was not the case for plants and the abiotic variables (Mantel r = 0.060, p = .318; Figure S6d-f).

| DISCUSSION
In our study, we compared the PSF effects of conditioning of the soil  pathogenic fungi and nematodes, have been suggested to maintain plant diversity at small spatial scales (e.g. Bever et al., 2015;De Kroon et al., 2012;Petermann et al., 2008). Our data now add to this by showing that the diversity of the plant species conditioning the soil drives the evenness of the plant community subsequently growing in that soil, through PSF. As low community evenness translates into higher local extinction rates (Hillebrand et al., 2008;Wilsey & Polley, 2004), we propose that these differences can result in differences in species richness over longer time frames.
We found that direct PSFs in uniform mono-conditioned soils were often negative and strong enough to alter competitive hierarchies (Hendriks et al., 2013;Jing, Bezemer, & Van der Putten, 2015;Pendergast, Burke, & Carson, 2013;Shannon, Flory, & Reynolds, 2012). This led to heterospecifics typically becoming the most abundant in the community. In soils conditioned by multiple species, PSF effects were less pronounced. In addition, the range of competitive abilities among species was smaller on mixed-conditioned soils, and this led to more equalized shoot biomass production among species and thus higher community evenness. Although we cannot demonstrate this directly here, our data underscore the idea that PSFs, through their frequency-dependent effects on plant performance, may mediate competitive intransitivity among species (De Kroon & Jongejans, 2016;De Kroon et al., 2012;Soliveres et al., 2015), which is thought to influence dynamics of many natural communities (Soliveres et al., 2015). Alternatively, PSFs may only lead to equalized competitive abilities, which would falsify PSF as stabilizing mechanism of plant diversity. As a next step, pairwise competition experiments among all species in the community are needed to definitively demonstrate PSFmediated intransitivity of competitive abilities (Jing et al., 2015;Laird & Schamp, 2006;Petraitis, 1979).
We found strong direct PSF effects for most plant species, in shoot biomass and also mortality, as was found in other studies (Kardol, Cornips, van Kempen, Bakx-Schotman, & Van der Putten, 2007;Kulmatiski et al., 2008;Petermann et al., 2008). However, PSFs were not strong enough to prevent all species from becoming dominant in their own self-conditioned soils. Both F. rubra and H. radicata were the F I G U R E 4 Relative abundance (M ± SE; shoot biomass) per species and conditioned soil in two plant mixtures in mono-conditioned uniform (a, b) and two heterogeneous, coarse (c, d)-and fine-grained (e, f), PSF treatments. Shown are all four plant species (coloured bars) per community on all four conditioned soils (coloured squares below bars) in that mix (Figure 1). The relative abundances sum to 1 per panel (i.e. all test species × soil patch combination per level of spatial heterogeneity). Hatched bars indicate that plants grew on self-conditioned soil. Different letters indicate significant differences among bars and those differences were tested separately per species and level of spatial heterogeneity. For abbreviations, see Figure 1, and for statistical analysis, see Table 1. For comparison, the same analyses were also conducted on shoot biomass; see Figure S3 and Table S5 Relative   (Brandt et al., 2013;Burns & Brandt, 2014;Grubb, 1977), which we did not test as the experiment started from seedlings. If PSF effects are stronger during early life stages (Kardol, De Deyn, Laliberté, Mariotte, & Hawkes, 2013), PSF effects could therefore be stronger in the field than reported here, but this is insufficiently studied so far.
A long-standing alternative explanation for local plant diversity and evenness has been that spatially heterogeneous abiotic conditions, such as soil nutrient availability, create niches that can be occupied by different species (Harpole et al., 2016;Tilman, 1982). However, metaanalyses testing the effects of spatial heterogeneity in abiotic conditions on plant diversity have shown that abiotic heterogeneity only has a positive effect on local diversity at scales of heterogeneity (i.e. grain) that exceed the reach of interacting plants (sensu Casper et al., 2003).
Plant-soil feedbacks can be mediated both abiotically and by the soil community (Ehrenfeld, Ravit, & Elgersma, 2005 Bradley et al., 2008;Hendriks et al., 2013), and community evenness. In general, our results support the notion that at small spatial scales, soil biota, not abiotic spatial heterogeneity, drives local plant diversity by preventing competitive exclusion De Kroon et al., 2012;Petermann et al., 2008).
Furthermore, plant evenness in pots with soil simultaneously conditioned by four species (mixed-conditioned uniform) was similar to that in the spatially heterogeneous pots but higher than in the monoconditioned uniform treatment. It is important to highlight, however, that there might have been a difference in host-microbe interactions during soil conditioning. During conditioning in mono-conditioned soils, the host plant relative abundance was high (i.e. host relative abundance 4/4), compared to mixed-conditioning (i.e. host relative abundance 1/4), and this may have led to more pronounced effects on soil community composition. In future studies, mono-conditioned soils need to be mixed and tested alongside mixed-conditioned soils to tease apart the influence of plant relative abundance on host-microbe interactions during the conditioning phase. Nevertheless, in line with our results, Hendriks et al. (2013) showed that mixing own and foreign soil releases plants from their negative self-feedback. Models of PSFmediated plant coexistence suggest that PSF effects need to be highly localized to maintain diversity (Abbott et al., 2015;Bonanomi et al., 2005;Fukami & Nakajima, 2013;Mack & Bever, 2014). Our study was carried out at a scale where the roots of all plant individuals could forage the entire experimental unit and be in contact with all soils (i.e. there was no physical barrier between patches) independent of the spatial configuration. Indeed, we observed that root systems spread through the entire container. Altogether, our results suggest that as long as multiple species conditioned the soil within the plant roots' zone of influence (Casper et al., 2003), the exact spatial pattern of conditioning is less important.  (Bezemer et al., 2010), but how they build up over the lifetime of a plant (Kardol et al., 2013;Zhang, Van der Putten, & Veen, 2016) and whether these induce the same PSF effects as in the glasshouse is unclear. In addition, It will be key to investigate how important PSFs are in driving plant community composition in relation to other mechanisms, for example the colonization-competition trade-off (Tilman, 1994), as well as in interaction with large herbivores (Chesson & Kuang, 2008;Veen et al., 2014) and across abiotic gradients .
In conclusion, we show that PSFs promoted plant community evenness when multiple species conditioned the soil, but that at small spatial scales, the spatial distribution of PSFs did not significantly affect plant community evenness. The presence of soil conditioned by all plant species in the community lead to more equal competitive abilities among plant species relative to soil conditioned by a single species. The spatial grain of PSF heterogeneity had no strong effect, suggesting that it is the presence, in sufficient amount, of each species' soil-borne antagonists that promotes plant evenness. Future studies are needed to quantify the importance of PSF in the field relative to other environmental factors and across plant life stages and generations.
Thanks also for suggestions on interpretation from Gera Hol, Wim Van