Diversity-dependent plant–soil feedbacks underlie long-term plant diversity effects on primary productivity
Abstract
Although diversity-dependent plant–soil feedbacks (PSFs) may contribute significantly to plant diversity effects on ecosystem functioning, the influences of underlying abiotic and biotic mechanistic pathways have been little explored to date. Here, we assessed such pathways with a PSF experiment using soil conditioned for ≥12 yr from two grassland biodiversity experiments. Model plant communities differing in plant species and functional group richness (current plant diversity treatment) were grown in soils conditioned by plant communities with either low- or high-diversity (soil history treatment). Our results indicate that plant diversity can modify plant productivity through both diversity-mediated plant–plant and plant–soil interactions, with the main driver (current plant diversity or soil history) differing with experimental context. Structural equation modeling suggests that the underlying mechanisms of PSFs were explained to a significant extent by both abiotic and biotic pathways (specifically, soil nitrogen availability and soil nematode richness). Thus, effects of plant diversity loss on plant productivity may persist or even increase over time because of biotic and abiotic soil legacy effects.
Introduction
Increasingly positive plant diversity effects on ecosystem functioning over time have been found in experimental ecosystems (Cardinale et al. 2007, Reich et al. 2012, Meyer et al. 2016, Guerrero-Ramírez et al. 2017). Temporal strengthening of the biodiversity–ecosystem functioning (B-EF) relationship can be driven, depending on the experimental context, for example, soil conditions, by an increase in primary productivity in high-diversity communities, a decrease in primary productivity in low-diversity communities, or both (Guerrero-Ramírez et al. 2017). Therefore, temporal changes in the B-EF relationship depend not only on mechanisms underlying an increase of ecosystem functioning in communities with high-diversity, which has been the main focus in previous work (Cardinale et al. 2007, Fargione et al. 2007, Reich et al. 2012), but also on mechanisms underlying a decrease of plant primary productivity in communities with low-diversity (Maron et al. 2011, Schnitzer et al. 2011).
A strengthening of the relationship between plant diversity and primary productivity may result from dynamic temporal interactions between plants and their soil biotic and abiotic environment. Specifically, plant communities differing in their diversity may increase or decrease their productivity over time by modifying their soil environment, that is, inducing positive or negative plant–soil feedbacks (PSFs; Eisenhauer et al. 2012a, Kulmatiski et al. 2012, van der Putten et al. 2013). Biotic and abiotic drivers of PSFs have been widely studied, for example, effects of plant species and functional group identity (Bezemer et al. 2006), plant community successional stage (van der Putten et al. 1993, Kardol et al. 2006), and plant traits (Baxendale et al. 2014, Bennett et al. 2017, Teste et al. 2017). Yet, our understanding of the role of PSFs in biodiversity–ecosystem functioning relationships remains limited (van der Putten et al. 2013).
Positive PSFs may likely be the result of high-diversity plant communities allowing for the diversification of soil microhabitats that, in turn, likely promotes the diversity of soil organisms (Hooper et al. 2000, Eisenhauer et al. 2017). High-diversity plant communities may also accumulate and exploit more soil nutrients than less diverse communities, for example, through plant resource use complementarity (Loreau and Hector 2001, Tilman et al. 2014), creating a resource gradient with increasing plant diversity over time (Fargione et al. 2007, Oelmann et al. 2010, 2011, Lange et al. 2015). These changes in the soil biotic and abiotic environment may, in turn, feedback on primary productivity (Bever et al. 2010). For example, high-diversity soil communities may increase primary productivity by influencing resource partitioning via increasing the available biotope space (Dimitrakopoulos and Schmid 2004, Eisenhauer et al. 2012b), and the available forms of soil nutrients (Turner 2008, Eisenhauer 2012).
Alternatively, negative PSFs may be the result of low-diversity plant communities being dominated by plant antagonists, for example, due to missing dilution effects of host plants (Civitello et al. 2015) and/or similar plant defenses (van der Putten 2003). Low-diversity plant communities may also use soil nutrients in an imbalanced way and thus deplete soil nutrient stocks (Schenk 2006) and/or be less efficient in nutrient retention (Leimer et al. 2016). Thus, soil conditioned by low-diversity plant communities may have a negative effect on plant productivity in the long run (Yang et al. 2015), and thereby co-determine the biodiversity–productivity relationship when used as a baseline to calculate diversity effects on ecosystem functioning.
Plant–soil feedbacks have been determined mainly using responses at the species level and, to a lesser extent, responses at the community level. Evidence suggests that the direction and magnitude of PSFs may differ between levels of organization, with a stronger negative PSF at the species level as compared to the community level (Kulmatiski et al. 2008). However, caution is required due to the lack of available information at the community level. In the context of the biodiversity–productivity relationship, it is likely that a positive PSF may influence responses particularly at the community level. Specifically, potential PSFs of soil conditioned by high-diversity plant communities, for example, via increases in soil biodiversity and available nutrient forms, may be exploited more completely by diverse plant communities (by more species and/or functional groups), for example, due to resource use complementarity (Loreau and Hector 2001, Tilman et al. 2014), and not by low-diversity communities.
The impact of low-diversity plant communities on plant productivity via soil may differ based on the functional relatedness between the plant community that conditioned and the plant community currently growing in the soil (Bezemer et al. 2006, Cortois et al. 2016). Similar plant functional traits and close relatedness may amplify negative PSFs due to the accumulation of specialized plant antagonist and/or depleted specific nutrient stocks. Alternatively, similar plant functional traits and close relatedness may also amplify positive PSFs due to the accumulation of specialized plant facilitators (Cortois et al. 2016). Therefore, the result of the degree of specialization of both, antagonist and facilitators, and the strategies to counteract generalist antagonists, for example, differences among plant functional groups in plant defenses (van der Putten 2003) or plant palatability (Wallinger et al. 2014), and/or benefit from generalist plant growth facilitators, for example, unequal competitors for nutrients in the soil (Tilman and Wedin 1991), may play a key role in PSFs.
Here, we aimed to assess the role of PSFs in plant diversity effects on primary productivity in two long-term plant diversity experiments. These two long-term plant diversity experiments were shown to differ in their temporal trajectory of biodiversity–productivity relationships, with soil conditions not only affecting these temporal patterns (Guerrero-Ramírez et al. 2017), but likely also the mechanism underlying these patterns, such as PSFs. To assess the role of PSFs, we determined these using soil conditioned for more than twelve years by plant communities with either low- or high-diversity (soil history). To consider multiple potential mechanistic pathways that underlie the effects of PSFs on plant productivity, we explored the roles of soil biotic (soil nematode diversity and functional composition) and abiotic (available soil nutrients) conditions.
While the role of soil microorganisms in biodiversity–productivity relationships has been shown before (Maron et al. 2011, Schnitzer et al. 2011, Wagg et al. 2011a, b, Latz et al. 2012), we here focused on nematode communities. Nematodes represent the most abundant metazoans on Earth and are likely to exert various effects on plant growth, potentially providing a mechanistic explanation of PSFs (van der Putten et al. 1993, Cortois et al. 2017). Indeed, previous work in grassland plant diversity experiments has shown that the functional composition of soil nematode communities is significantly affected by plant diversity (De Deyn et al. 2004, Viketoft et al. 2009, Eisenhauer et al. 2011), with likely feedback effects on plant growth and subsequently biodiversity–productivity relationships (Eisenhauer et al. 2011). Therefore, we hypothesized positive effects of soil conditioned by high-diversity plant communities on plant productivity compared with soil conditioned by low-diversity plant communities. We expected contrasting plant diversity-induced soil legacy effects on plant productivity to act via dissimilar biotic and abiotic soil conditions. Specifically, on the one hand, we expected soil conditioned by high-diversity plant communities to be associated with high nematode richness, high density of plant growth facilitating nematodes (i.e., all functional trophic guilds except plant-feeders; Kulmatiski et al. 2014), and high availability of soil nutrients. On the other hand, we expected soil conditioned by low-diversity plant communities to be associated with high abundance of plant-feeding nematodes and a reduced availability of nutrients. Moreover, we expected that the positive PSF of soil conditioned by high-diversity plant communities to be stronger in plant species mixtures in the feedback phase than in plant communities with only one plant functional group. Finally, we hypothesized the negative PSF of soil conditioned by low-diversity plant communities to be stronger in low-diversity plant communities in the feedback phase growing in soil conditioned by their own plant functional group than in soil conditioned by other plant functional groups.
Materials and Methods
Conditioning phase
Soil from two long-term grassland biodiversity experiments, the BioCON Experiment (Reich et al. 2001) and the Jena Experiment (Roscher et al. 2004), was used to establish a PSF experiment. The BioCON Experiment is located at the Cedar Creek Ecosystem Science Reserve in Minnesota, USA, on sandy soil (Nymore series, Typic Upidsamment). The BioCON Experiment was planted in 1997 using a pool of 16 species including grasses, forbs, and legumes (Reich et al. 2001). The Jena Experiment is located in Jena, Germany, on the floodplain of the River Saale, and the soil is Eutric Fluvisol. The Jena Experiment was sown in 2002 using a pool of 60 species including grasses, forbs, and legumes (Roscher et al. 2004).
From each biodiversity experiment, soil samples from 12 plots with one plant species (low-diversity plots) and 5 plots with either nine (BioCON Experiment) or eight (Jena Experiment) plant species (high-diversity plots) were collected in summer of 2014. Soil was conditioned for 17 and 12 yr in the BioCON Experiment and the Jena Experiment, respectively. In the BioCON Experiment, we only sampled plots with ambient atmospheric CO2 concentrations and N availability. For low-diversity plots, a minimum of three plots was selected for each plant functional group: grasses, forbs, and legumes. For high-diversity plots, all selected plots contained the three plant functional groups together. For both, low- and high-diversity plots, the variability within plant functional groups presented in each biodiversity experiment was accounted by incorporating both, C3 and C4 grasses, in the case of the BioCON Experiment (grasses represented 50% of the species pool in the BioCON Experiment; Reich et al. 2001) and both, short and tall herbaceous species, in the case of the Jena Experiment (forbs represented ~50% of the species pool in the Jena Experiment; Roscher et al. 2004). In each plot, three soil cores were taken to cover the spatial heterogeneity of the plots (depth 20 cm, diameter 3.8 cm in the BioCON Experiment or 5 cm in the Jena Experiment) and mixed gently in a plastic bag (between 600 and 800 g of fresh soil were collected per plot, respectively). Afterward, the soil was carefully sieved using a 4-mm mesh to remove rocks and roots.
From each soil sample, a sub-sample of 25 g of fresh soil was taken to extract nematodes using a modified Baermann method (Ruess 1995; Appendix S1). This amount of soil was shown to have high nematode extraction efficiency (Schulz et al. 2018). Nematodes were then grouped into different feeding groups (plant-feeders, bacterial-feeders, fungal-feeders, omnivores, and predators), and their abundances were assessed. This information was used in a principal component analysis (PCA) to determine variation in the functional composition of nematode communities among experimental plots (Appendix S7: Figs. S1, S2). PC axis two was used in the statistical analyses, because it represents the potential negative role of plant-feeders in both biodiversity experiments.
Plant available nitrogen (N) and phosphorus (P) were also quantified using sub-samples taken from the soil in the two biodiversity experiments. A proxy of extractable or available N (mg/kg) was calculated by adding the N concentrations of ammonium (NH4) and nitrate (NO3; Appendix S1).
Feedback phase
Soil collected from each plot was divided into four sub-samples to establish four new plant communities in a microcosm experiment: one plant community for each plant functional group (grasses, forbs, and legumes) and one plant community that contained all three plant functional groups (Fig. 1a). The feedback experiment was established in microcosms (diameter of 6 cm at the bottom and 9 cm at the top, and a height of 7 cm; Appendix S2). Two plant species for each plant functional group were selected for each experimental site. To avoid species-specific PSFs, the selected plant species used in the feedback experiment were present in the study area but were not present in the sampled plots (Appendix S2). Each microcosm contained six plant individuals to keep plant density constant among treatments. After transplanting, the experiment was run for six weeks.

In total, 136 microcosms were established: two studies (the BioCON Experiment and the Jena Experiment) × 17 plots (12 soils conditioned by low- and five soils conditioned by high-diversity plant communities, i.e., soil history) × 4 plant communities (grasses only, forbs only, legumes only, and a mixture of all three plant functional groups). At the end of the experiment, shoot and root biomass was harvested and dried at 70°C for 72 h and weighed (0.001 g).
To determine plant diversity effects, proportional deviance was calculated (Loreau 1998; Appendix S3). To test whether non-additive diversity effects were found in plant mixtures, 95% confidence intervals (CI) were calculated using 10,000 bootstrap replicates (adjusted percentile bootstrap). Non-additive effects are considered to occur when the 95% CI do not overlap with zero.
Based on Baxendale et al. (2014), a PSF index was calculated to determine the role of plant functional identity of low-diversity communities (Appendix S4). To test whether plant functional identity affected PSFs, 95% CI were calculated using 10,000 bootstrap replicates (adjusted percentile bootstrap). Positive values indicated higher plant biomass in plant communities growing in soil conditioned by their own plant functional group than in soil conditioned by other plant functional groups. In contrast, negative values indicated higher plant biomass in plant communities growing in soil conditioned by other plant functional groups than in soil conditioned by their own plant functional group. Strong (positive or negative) PSFs are considered to occur when the 95% CI do not overlap with zero.
Statistical analysis
The effects of soil history (conditioning phase), plant diversity in microcosms (feedback phase), and their interaction on total, shoot, and root biomass of plant communities in the microcosm experiment were determined with a linear mixed-effect model using the nlme package (Pinheiro et al. 2018). To account for potential heterogeneity in soil conditions (Reich et al. 2001, Roscher et al. 2004), plot nested in experimental blocks/rings was included as a random effect (four blocks and three rings for the Jena Experiment and BioCON Experiment, respectively). To fulfill the assumptions of linear mixed-effect models, square-root transformations were used in the BioCON Experiment (Appendix S5).
To assess the influence of soil history on plant biomass operating via soil biotic and abiotic conditions, structural equation modeling (SEM) based on piecewise fitting of linear mixed-effect models using the piecewiseSEM package was performed (Lefcheck 2016). The SEM allowed us to test a hypothetical causal model based on a priori knowledge of PSFs (Fig. 1b). A direct path was included between soil history (plant diversity levels in the conditioning phase) and nematode communities, that is, nematodes richness and functional composition (PC axis 2), and between soil history and soil available nutrients (N and P concentrations). As nematode communities and soil available nutrients are potential mechanistic pathways explaining PSFs, alternative paths between them and plant biomass were added, if this improved the model fit (based on modification indices, P-value < 0.05). Effects of soil history through mechanistic pathways were calculated by multiplying the effect of soil history on the biotic/abiotic explanatory variable and the effect of the biotic/abiotic explanatory variable on plant biomass in the microcosms. Mechanisms that were not captured by neither of the mechanistic pathways are represented by the direct path between soil history and plant biomass of plant communities in the microcosm experiment. Experimental blocks/rings were also included as a random effect. Independent path models were fitted for each biodiversity experiment (the BioCON Experiment and the Jena Experiment). To fulfill the assumptions of the models in the BioCON Experiment and the Jena Experiment, natural logarithmic and square-root transformations were carried out (Appendix S5). Model fits were assessed using Fisher's C statistic based on the test of directed separation; if the test P-value was >0.05, the data fit the hypothesized causal network (Lefcheck 2016). All analyses were performed using R 3.4.1 (R Core Team 2018).
Results
Effects of soil history, plant diversity, and their interaction on plant biomass
Soil history (conditioning phase) and the interaction between soil history and plant diversity in the microcosm experiment (feedback phase) influenced plant biomass in the BioCON Experiment (Appendix S7: Table S1; Fig. 2). Plant biomass was higher in plant communities that grew in soil conditioned by high-diversity plant communities (F1,13 = 6.059 and 6.316 for total [+41%] and shoot biomass [+54%], respectively, P-values < 0.05; F1,13 = 3.450 for root biomass [+29%], P-value < 0.1). The positive effect of soil history on plant biomass in microcosms was stronger in plant mixtures than in plant communities with a single plant functional group (soil history × plant diversity in microcosms: F1,48 = 3.179 and 3.367 for total [+155% in plant mixtures and +19% in single plant functional groups] and shoot biomass [+169% in plant mixtures and +25% single plant functional groups], respectively, P-values < 0.1).

In the Jena Experiment, soil history did not significantly influence plant biomass in the microcosm experiment (Appendix S7: Table S1). Instead, plant biomass increased with current plant diversity in microcosms only (F1,49 = 4.235 and 5.175 for total [+49%] and root biomass [+47%], respectively, P-values < 0.05; F1,49 = 4.235 for shoot biomass [+51%], P-value < 0.1; Fig. 2).
Effects of soil history on total plant biomass through differential mechanistic pathways
Plant–soil feedbacks played a key role in determining total plant biomass in mixtures in the BioCON Experiment (Appendix S7: Table S2; Fig. 3a) via two mechanistic pathways: nematode richness and soil available N (Appendix S7: Table S2; Fig. 3a). Nematode richness and soil available N were higher in soils conditioned by high- than by low-diversity plant communities, with higher nematode richness and available N increasing total plant biomass in plant mixtures in microcosms (+115%; standardized effects via nematode richness = 0.28 and via available N = 0.20). Plant–soil feedbacks in the BioCON Experiment were not fully explained by the measured mechanistic pathways, with a remaining significant path between soil history and total plant biomass in mixtures (standardized effect = 0.23). Soil available P affected total biomass in plant mixtures in microcosms in the BioCON Experiment and the Jena Experiment (Appendix S7: Tables S2, S3; Fig. 3a). Yet, the effect of available P (standardized effects = −0.16 and 0.74 in the BioCON Experiment and the Jena Experiment, respectively) was not related to PSFs associated with plant diversity in the conditioning phase.

Plant–soil feedbacks played a key role in determining total plant biomass of single plant functional groups in the BioCON Experiment, with higher total plant biomass of forbs and legumes growing alone in the microcosms in soil conditioned by high-diversity plant communities (+45% and +8%, respectively). For total biomass of forbs, neither of the mechanistic pathways explained PSFs (Appendix S7: Table S4; Fig. 3b). Thus, PSFs were explained by a significant path between soil history and total plant biomass of forbs (standardized effect = 0.56). For total biomass of legumes, PSFs acted via available N (standardized effect = 0.74). However, PSFs for legumes growing alone were not fully explained by the measured mechanistic pathways, with a significant negative path between soil history and total biomass (standardized effect = −0.67). Plant–soil feedbacks did not influence total plant biomass of grasses growing alone in the microcosm experiment (Appendix S7: Table S4; Fig. 3b).
In the Jena Experiment, PSFs affected total plant biomass of single plant functional groups, specifically grasses and legumes growing alone in the microcosms. Explicitly, total biomass of grasses and legumes was higher in communities growing in soil conditioned by high- than by low-diversity communities (+23% and +13%, respectively). Yet, neither of the measured mechanistic pathways explained the PSFs (Appendix S7: Table S5; Fig. 3b). Thus, there was a significant positive pathway between soil history and total biomass of grasses or legumes (standardized effect = 0.54 or 0.39, respectively). Plant–soil feedbacks did not significantly influence total biomass of forbs.
The functional composition of nematode communities affected total biomass of forbs growing alone in microcosms in the BioCON Experiment and in the Jena Experiment (Appendix S7: Tables S4, S5; Fig. 3b). Yet, their influence was not related to PSFs induced by plant diversity in the conditioning phase. In the BioCON Experiment, soils with a higher abundance of fungal-feeding and lower abundance of plant-feeding nematodes increased the total biomass of forbs (standardized effect = 0.54). In the Jena Experiment, soils with higher abundance of plant-feeders and omnivores as well as a lower abundance of fungal-feeding nematodes increased the total biomass of forbs (standardized effect = 0.56). In the Jena Experiment, available P affected total biomass of legumes growing alone, but this effect was not related to PSFs induced by plant diversity in the conditioning phase (standardized effect = 0.68).
Detailed results of effects of soil history on plant shoot and root biomass allocation are present in Appendix S6.
Effects of soil history on plant biomass: the role of plant functional group identity
In the BioCON Experiment, negative values of the PSF index indicated that plant communities with a single plant functional group in the microcosms grew significantly worse (for total and root biomass) in soil conditioned by low-diversity communities with the same plant functional group as compared to low-diversity communities conditioned by other plant functional groups (Fig. 4a). In the Jena Experiment, a similar pattern was observed, but it was not statistically significant, that is, 95% CI overlapped zero.

Effects of soil history on plant diversity effects on plant biomass
Soil history changed not only plant biomass per se but also influenced the strength of plant diversity effects in the feedback phase in the BioCON Experiment (Fig. 4b). Notably, positive plant diversity effects on total and shoot biomass were observed when plant communities grew in soil conditioned by high- but not by low-diversity plant communities (i.e., 95% CI overlapped zero). In the Jena Experiment, positive plant diversity effects in the feedback phase on total, shoot, and root biomass were observed independent of the soil history.
Discussion
This is one of the first studies that disentangles potential mechanistic pathways underlying diversity-dependent PSFs on primary productivity and plant diversity effects in long-term biodiversity experiments. Our results indicate that PSFs may drive positive plant diversity effects on ecosystem functioning via abiotic and biotic changes in soil conditions.
Positive PSFs of soil conditioned by high-diversity plant communities in the BioCON Experiment mainly influenced diverse plant communities and not communities with a single plant functional group in the feedback phase. Therefore, PSFs in the BioCON Experiment not only affected plant productivity but also explained positive plant diversity effects in our microcosm experiment. Moreover, low-diversity communities were negatively affected by growing in soil conditioned by their own plant functional group, especially in the BioCON Experiment, suggesting that negative effects associated with low-diversity communities may be amplified when plant species are functionally similar. In contrast, plant–plant interactions, and not plant–soil interactions, drove diversity effects on primary productivity in the Jena Experiment. Taken together, these results suggest that primary productivity and plant diversity effects on this ecosystem function can be driven by plant–soil and plant–plant interactions, with the main driver differing between experimental contexts.
Positive PSFs on plant productivity emerged due to the influence of plant diversity of both the community conditioning the soil and the community growing in the conditioned soil. The absence of positive PSFs of high-diversity plant communities in other studies (Yang et al. 2015) may therefore sometimes be the result of assessing PSFs at the species level and not at the community level, such as done in the present study. High-diversity communities may positively affect plant productivity by diversifying the abiotic and biotic biotope characteristics that facilitate plant growth (Hooper et al. 2000, Eisenhauer et al. 2012b). Yet, potential mechanisms required to capitalize on this positive PSF may be present mostly in high-diversity communities and not in low-diversity communities (Loreau and Hector 2001, Tilman et al. 2014).
The mechanisms underlying diversity-dependent PSFs on primary productivity are likely interconnected and dynamic, increasing the challenge of disentangling confounding factors to establish causality. For instance, plant diversity of the soil conditioning community increased productivity via available N. Although nutrient depletion (in low-diversity communities) and/or enrichment (in high-diversity communities) may explain the observed PSFs, other potential mechanisms are also possible. For instance, a decrease in root biomass has been suggested as a potential mechanism explaining a temporal decrease in available N (Mueller et al. 2013). Pathogens could exacerbate the decrease of root biomass, thereby reducing either the recruitment or the performance of specific plant species or functional groups (Fornara and Tilman 2008, Petermann et al. 2008, Reich et al. 2012).
Inconsistent PSFs on primary productivity between biodiversity experiments may reflect differences in the environmental context, such as soil fertility (Guerrero-Ramírez et al. 2017). Positive effects of plant diversity on available N may be more pronounced in soils with low fertility, such as the sandy and organic-poor soil in the BioCON Experiment. Positive effects of plant diversity on NH4 may emerge because of an increase of soil carbon concentration over time in more diverse communities (Fornara and Tilman 2008, Mueller et al. 2013). Temporal dynamics of nutrients can be also observed in more fertile soils as in the Jena Experiment (Oelmann et al. 2011). However, weaker effects may emerge due to either constraint in the potential of plant diversity to significantly increase nutrient availability in fertile soils or the absence of specific mechanisms driving the contrasting effects of high- and low-diversity plant communities on available nutrients, for example, differences in root biomass across plant richness levels (Bessler et al. 2009). It is also possible that PSFs in more fertile soils may not affect plant productivity per se but rather the recruitment of specific species (LaManna et al. 2016). Also, changes on plant productivity may be driven by other resources (Hautier et al. 2009) or by plant community evolutionary changes via selection (van Moorsel et al. 2018).
Dissimilarity of drivers of plant diversity effects on ecosystem functioning between biodiversity experiments may emerge as biotic interactions likely co-vary with environmental conditions. Biotic interactions can be impacted by soil chemical and physical characteristics that directly influence soil communities (Hassink et al. 1993, Wardle et al. 2004) and/or modify plant–microbial competition and facilitation (Kuzyakov and Xu 2013, Johnson and Rasmann 2015). The fertility of the ecosystem may shape plant community composition as well as impact energy channels, with these changes cascading to ecosystem processes (Wardle et al. 2004). Positive effects of plant growth facilitators are often more pronounced in nutrient-poor than in nutrient-rich soils (Johnson 2009, van Groenigen et al. 2014). Differences in soil characteristics like soil texture affect the grazing activity of nematodes (bacterial-feeders), resulting in higher soil N mineralization rates in sandy than in clay soils (Hassink et al. 1993). Further, effects of plant antagonists may differ between soil fertility levels through changes in compensatory plant growth responses to herbivory. A reduction of compensatory effects in nutrient-poor soils may amplify the negative effects of herbivory, affecting not only plant growth but also herbivore–plant–microbial interactions (Bardgett and Wardle 2003). Thus, soil fertility may co-determine plant–soil biotic interactions as well as the consequences for ecosystem functioning.
It is also likely than other intrinsic abiotic and biotic differences between the two plant diversity experiments contributed to the observed context dependency. For instance, Fargione et al. (2007) reported that the presence of both, C4 grasses and legumes combined, contribute to the accumulation of N in more diverse communities with this positive effect increasing over time. Yet, C4 grasses are only present in the BioCON Experiment but not in the Jena Experiment.
In conclusion, temporal changes in the biodiversity–productivity relationship may emerge because of diversity-dependent PSFs. However, diversity-dependent PSFs rely on the experimental context, reinforcing the notion that environmental conditions are likely to play a key role in determining not only the patterns (Guerrero-Ramírez et al. 2017), but also the mechanisms underlying temporal changes in the biodiversity–productivity relationship in grasslands. To our knowledge, this is one of the first studies disentangling diversity-dependent PSFs, and further studies are required. For instance, long-term biodiversity experiments along environmental gradients are necessary to assess potential drivers involved in diversity-dependent PSFs. Since there is only a handful of long-term grassland plant diversity experiments existing worldwide to date, these are of particular value to improve our mechanistic understanding. Moreover, in our study, it is not possible to separate the role of plant functional groups from the role of plant species richness, requiring further experiments in this direction. From a mechanistic perspective, we encourage studies directly manipulating specific biotic (e.g., soil microbial or nematode communities) and abiotic (e.g., soil nutrient levels) drivers that are likely to contribute to diversity-dependent PSFs under different environmental contexts. Since plant diversity can modify ecosystem functioning through both plant–plant interactions as well as through soil legacy effects (Zuppinger-Dingley et al. 2016), negative effects of plant diversity loss on ecosystem functioning may persist or even increase over time.
Acknowledgments
We specially thank Alfred Lochner for his support during all the experimental phase and Petra Hoffmann for her support during the chemical analyses. We thank Kally Worm, Anne Ebeling, Simone Cesarz, Silke Schroeckh, Katja Steinauer, and the gardeners and coordinators of both biodiversity experiments. We thank Dylan Craven for providing comments on a previous version of this manuscript. We thank Andrew Barnes and Jon Lefcheck for their input on the structural equation models. This work was funded by the German Research Foundation (Deutsche Forschungsgemeinschaft, DFG) in the frame of the Emmy Noether research group (Ei 862/2) to Nico Eisenahuer and the German Centre for Integrative Biodiversity Research (iDiv) Halle-Jena-Leipzig, funded by the DFG (FZT 118). The Jena Experiment is financed by the DFG (FOR 1451). Support for the BioCON Experiment came from the U.S. National Science Foundation (NSF) Long-Term Ecological Research (DEB-9411972, DEB-0080382, DEB-0620652, and DEB-1234162), Biocomplexity Coupled Biogeochemical Cycles (DEB-0322057), Long-Term Research in Environmental Biology (DEB-0716587, DEB-1242531), and Ecosystem Sciences (NSF DEB-1120064) Programs; as well as the U.S. Department of Energy Programs for Ecosystem Research (vDE-FG02-96ER62291) and National Institute for Climatic Change Research (DE-FC02-06ER64158). Nico Eisenhauer and Nathaly R. Guerrero-Ramírez developed the idea, Peter B. Reich contributed with the first experimental phase, Nathaly R. Guerrero-Ramírez implemented the study and collected data with the help of Marcel Ciobanu and Cameron Wagg. Nathaly R. Guerrero-Ramírez analyzed the data and wrote the paper with substantial input from all authors.