Introduction

Anthropogenic global environmental change affects ecosystem properties worldwide and threatens important ecosystem functions1,2. Climate change is predicted to alter precipitation regimes towards more frequent and severe drought events in the future3. Simultaneously, human activities, such as fossil fuel combustion and fertilization, are causing an acceleration of the turnover rates of the nitrogen cycle and will double nitrogen deposition in the future4,5. The same is true for phosphorous inputs, which also increased at a global scale6. Thus, multiple global change drivers are occurring side by side, and their effects are not necessarily additive or antagonistic. Our knowledge on their interactive effects, however, is still highly limited7,8. This is particularly true for the responses of soil organisms, which mediate crucial ecosystem functions and services, such as nutrient cycling and decomposition9,10. Their significant role is not adequately reflected in the body of global change literature yet. Therefore, a more comprehensive understanding of above- and belowground dynamics is key to predict the responses of terrestrial ecosystems in a changing world7.

Many soil organisms are dependent on a water-saturated atmosphere or on water films on soil aggregates11,12,13,14. Altered precipitation patterns will result in drought periods, which are likely to have substantial effects on their abundances and community structure, thus affecting important soil organism-mediated ecosystem processes. Previous studies reported detrimental effects of drought on soil microbial respiration and biomass as well as a reduction of the diversity of microbial communities15. Furthermore, drought was shown to cause a decline in soil microarthropod abundances16. In contrast, drought seems to have only marginal effects on nematode community composition17. Yet, a reduction of soil moisture content can induce community shifts via lower trophic levels, often favouring fungal-feeding nematodes over bacteria-feeders, as fungi perform relatively better under dry conditions17,18.

Nutrient enrichment is another key factor that affects the soil community by altering the physical and chemical properties of the soil, e.g., by influencing pH, soil porosity, and organic fractions19,20,21. Nitrogen addition has been identified to decrease soil microbial respiration and biomass, often leading to shifts in the soil microbial community composition under the use of mineral fertilizer (NPK)22,23,24. On the other hand, fertilization treatments were shown to increase soil microbial catabolic and functional diversity25,26. Furthermore, nitrogen addition alters the nematode community structure towards bacterivores, thus promoting the bacterial-dominated decomposition pathway27, and was shown to simplify communities17. Similarly, nitrogen was reported to create unfavourable conditions for soil microarthropods, leading to declines in abundance as well as diversity28,29. At the same time, nitrogen enrichment is one of the major drivers determining aboveground primary production30. Nitrogen and phosphorous addition are known to increase total aboveground biomass and consequently the quantity and quality of plant litter input to the soil26,31. In addition, nitrogen also affects plant rhizodeposition by altering the amount and quality of substrates released from roots32,33. Both mechanisms can potentially enhance resource availability via bottom-up effects and can therefore increase soil microarthropod abundances34. Concurrently, the fertilization-induced increase in aboveground biomass may cause higher transpiration rates, which are likely to reinforce drought effects on soil organisms35.

To investigate the interactive effects of extreme drought events and fertilization (NPK), we established a field experiment at the UFZ Experimental Research Station (Bad Lauchstädt, Germany), which combines the treatments of two globally distributed networks – the Drought-Network and the Nutrient Network36. Here, we tested the responses of soil microorganisms, nematodes, and overall soil invertebrate activity to the interactive effects of extreme drought and fertilization (NPK) across all seasons. Based on prior research, we hypothesized that (1) drought will reduce the activity of soil organisms, whereas (2) fertilization will increase their activity, owing to enhanced plant litter input that subsequently increases resource availability for soil organisms. Furthermore, we predicted that (3) the interactive effects of drought and fertilization will result in detrimental conditions for soil organisms as the negative effects of drought were expected to be further enhanced by increased plant growth under fertilization, resulting in reduced soil water availability for soil organisms.

Methods

Research site

The study site is located at the Experimental Research Station of the Helmholtz Centre for Environmental Research (UFZ), which is situated in Bad Lauchstädt, Germany. The field site is located in the central German dry area with a mean annual precipitation of 487 mm and an average annual daily temperature of 8.9 °C (Meteorological data of Bad Lauchstädt, Helmholtz Centre for Environmental Research GmbH - UFZ, Department of Soil System Science, 1896–2017). The area represents an anthropogenic grassland, which is maintained by moderate mowing (twice a year since 2012). It is a successional plant community dominated by Vulpia myuros (L.) C. C. Gmel., Picris hieracioides (L.) and Taraxacum officinale (F. H. Wigg.) with Apera spica-venti (L.) P. Beauv. and Cirsium arvense (L.) Scop. being very common. The soil is classified as a haplic chernozem, developed upon carbonatic loess substrates, distinguished by a composition of 70% silt and 20% clay37. Within the upper 30 cm, the soil contains 0.18% total nitrogen, 1 g kg−1 total phosphorus, and 20 g kg−1 total potassium. For more details see Altermann et al.37.

Weather conditions

Weather conditions within the two-year sampling period of this study were in line with the long-term average (2005–2015) despite some exceptions (Meteorological data of Bad Lauchstädt, Helmholtz Centre for Environmental Research GmbH - UFZ, Department of Soil System Science): precipitation patterns deviated from the long-term average in 2016 with a dry May (21.2 mm compared to an average of 62.3 mm) and a wet June (80.2 mm compared to an average of 41.2). September tended to be drier than usual in both years (19.5 mm in 2016 and 22.1 mm in 2017 compared to an average of 51.8 mm).

Experimental design and treatments

The experimental site was established in March 2015. The experimental design consists of five blocks with five plots each. The plots have a size of 2 × 2 m and are arranged at a distance of 3 m from each other (Fig. S1). The experiment includes two treatments with two levels each (first applied in March 2016): drought (control/drought) and fertilization (no NPK/NPK addition), as well as their interaction (drought × fertilization). Notably, this experiment crosses treatments of two globally distributed experimental networks: the full NPK fertilization treatment of the Nutrient Network36 and the drought treatment of the Drought-Network (http://www.drought-net.colostate.edu/)38,39,40.

In order to simulate drought, a rainfall manipulation system was established39 using corrugated acrylic strips. The roofs have a size of 3 × 3 m and reduce precipitation by 55% throughout the year, simulating a severe long-term reduction in precipitation. Roofs were built with a slope of 20° to ensure water runoff and account for the expected snow load in the region. Exclusion of potential artefacts was realized by equal roof constructions using inverted acrylic strips intended to let rainfall pass41 (Fig. S2). To control for possible infrastructure effects of the roof constructions itself, a fifth plot was added to each block without any roof construction (ambient plots), thus receiving ambient precipitation (not crossed with the fertilization treatment and thus only used to assess if the roof construction itself affected soil water content, see Fig. S1). To validate the drought treatment, soil water content was quantified on all plots in every sampling campaign. All three precipitation levels differed significantly in their soil water content (Tukey’s HSD test, p < 0.05): as intended, the lowest soil water content was found for the drought treatment (−19.4% compared to the ambient plots). Also the infrastructure control plots (with concave roof constructions) differed significantly from the ambient plots (without roof construction), indicating that there were effects of the roof construction itself (−13.4%). Furthermore, soil water content varied significantly between seasons (Table S1; Fig. S3).

The fertilization treatment was realized by annual addition of a mixture of separate fertilizers for nitrogen (N), phosphorus (P) and potassium (K) (i.e. NPK fertilization; applied at 10 g m−2 y−1 by elemental mass for each of the three main fertilizer elements) before each growing season. Following the protocol of the Nutrient Network36, we used coated urea (CO(NH2)2) as nitrogen fertilizer (Multicote, Haifa – Pioneering the Future Haifa, Matam-Haifa, Israel), P2O5 as phosphorus fertilizer (Triple Super Phosphate, Delitzscher Landhandels- und Dienste GmbH, Delitzsch, Germany), and K2SO4 as potassium fertilizer (KaliSOP gran. max. 1.0% Cl, K + S KALI GmbH, Kassel, Germany). In addition, the micronutrient mix “Micromax Premium” (Everris, Geldermalsen, The Netherlands) was applied in the first treatment year36. Specifically, we applied 25 g m−² Multicote, 50.9 g m−² Triple Super Phosphate, and 22.3 g m−² KaliSOP per year, and 100 g m−² Micromax Premium in the first year.

Soil sampling

The first soil sampling took place in March 2016. Sampling campaigns were repeated every three months to cover every season (spring, summer, fall, winter) from March 2016 to December 2017 (i.e., eight samplings across two years). Samples were taken on all plots with roof construction (drought and control) with a steel core sampler (1 cm in diameter; 15 cm deep). Seven subsamples per plot were homogenized, sieved at 2 mm, and stored at 4 °C. Soil samples were used to determine soil water content and microbial respiration. In addition, nematodes were extracted from the soil samples in spring and summer of 2017, and pH was measured for all four seasons in 2017.

The Bait Lamina Test

Feeding activity of soil invertebrates was surveyed using the bait lamina test (Terra Protecta GmbH, Berlin, Germany), which presents a commonly used rapid ecosystem function assessment method42. The test uses rigid PVC sticks (1 mm × 6 mm × 120 mm) with 16 holes of 1.5 mm diameter in 5 mm distance. Original sticks were filled with a bait substrate consisting of 70% cellulose powder, 27% wheat bran, and 3% activated carbon, which was prepared according to the recommendations of Terra Protecta. The bait substrate is primarily consumed by mites, collembolans, nematodes, enchytraeids, millipedes, and earthworms, whereas microbial activity plays a minor role in bait loss43,44,45,46. The bait strips were inserted vertically into the soil with the topmost hole just below the ground surface. To avoid damaging the strips, a steel knife was used to prepare the ground prior to insertion. Five strips were used per plot to account for spatial heterogeneity47. For each sampling campaign, the bait lamina strips were removed from the soil after three weeks of exposure and evaluated directly in the field. Bait consumption was recorded for each of the 16 holes per strip as empty (1), partly empty (0.5), or filled (0). Thus, soil invertebrate feeding activity per bait strip could range from 0 to 16 (maximum feeding activity). Mean bait consumption per plot (averaged among the five strips) was calculated prior to statistical analyses.

Microbial biomass and activity

An O2-microcompensation system was used to measure the respiratory response of soil microorganisms in two separated steps using approximately 6.5 g of fresh soil48. First, basal respiration was determined as a measure of soil microbial activity (µl O2 h−1 g−1 soil dry weight) without the addition of any substrate. Second, the maximal respiratory response to a single addition of glucose (4 mg g−1 dry weight soil, solved in 1.5 ml distilled water) allowed us to determine soil microbial biomass (μg Cmic g−1 soil dry weight)49. For an overview of the experimental setup see Fig. S4.

Nematode analysis

Nematode extraction was conducted with a modified Baermann method50. Approximately 25 g of soil per plot were transferred to plastic vessels with a milk filter and a fine gauze (200 µm) at the bottom and placed in water-filled funnels. More water was added to saturate the soil samples and to ensure a connected water column throughout the sample and the funnel. Hence, nematodes migrated from the soil through the milk filter and the gauze into the water column and gravitationally-settled at the bottom of a closed tube connected to the funnel. After 72 h at 20 °C, the nematodes were transferred to a 4% formaldehyde solution. Nematodes were counted at 100x magnification using a Leica DMI 4000B light microscope. Identification was conducted at 400x magnification. For identification, sediment material from the bottom of each sample vial was extracted with a 2 ml plastic pipette and examined in temporary mounted microscope slides. At least 100 well-preserved specimens (if available in the sample) were randomly selected and identified to genus (adults and most of the juveniles) or family level (juveniles), following Bongers (1988)51. Nematode taxa were then arranged into trophic groups (bacteria-, fungal- and plant-feeders, omnivores and predators)52,53. Due to low densities, omnivorous and predatory nematodes were grouped into a combined feeding type for most analyses, which was based on the fact that both groups often share similar traits, like being carnivores (i.e. representing higher trophic levels) and being persisters rather than colonizers on the c-p scale (see explanation below), thus responding similarly to environmental disturbances54. Nematodes were ordered according to the colonization-persistence gradient (c-p values)55,56, which classifies nematode taxa based on their life history strategy (i.e. r or K strategists). Cp-1 taxa are distinguished by their short generation cycles and high fecundity. They mainly feed on bacteria. Cp-2 taxa have longer generation times, lower fecundity and consist of bacterivores and fungivores57. Both are categorized as r-strategists. Cp-3 to cp-5 are classified as K-strategist nematodes with longer generation times, higher trophic feeding levels and increasing sensitivity against disturbances57. The c-p-values can be used to calculate the Maturity Index (MI) as weighted means of nematode families assigned to c-p-values. It is used to describe soil health and as an indicator of overall food web complexity55,56.

$$MI=\sum _{i=1}^{n}\,(\begin{array}{c}n\\ k\end{array})\,v(i)\,\ast \,f(i)$$

with v(i) being the c-p-value of a taxon i and f(i) being the frequency of that taxon in a sample.

Furthermore, nematode taxa were assigned to functional guilds according to Ferris et al.57, which then served as a basis to calculate additional indices. Functional guilds refer to the following trophic groups: bacterial feeders (BaX), fungal feeders (FuX), omnivores (OmX), and carnivores (CaX). Associated numbers (i.e., the x of the respective trophic group) are again referring to the c-p values described above. The Enrichment Index (EI) indicates the responsiveness of the opportunistic bacterial (Ba1 and Ba2) and fungal feeders (Fu2) to food web enrichment57 and is calculated as follows:

$${\rm{EI}}=100\times [\frac{{\rm{e}}}{{\rm{e}}+{\rm{b}}}]$$

with e as weighted frequencies of Ba1 and Fu2 and b as weighted frequencies of Ba2 and Fu2 nematodes57. The Channel Index (CI) reflects the nature of decomposition channels through the soil food web. High values indicate a predominant decomposition pathway of organic matter dominated by fungal-feeding nematodes, whereas low values refer to bacterial-dominated decomposition pathways57.

$${\rm{CI}}=100\times [0.8\times \frac{{\rm{Fu}}2}{3.2\times {\rm{Ba}}1+0.8\times {\rm{Fu}}2}]$$

with 0.8 and 3.2 representing enrichment weightings for Fu2 and Ba1 nematodes57. The Structure Index (SI) provides information about the complexity of the soil food web. A highly structured food web with a high SI suggests ecosystem stability, while low values imply environmental disturbance57.

$${\rm{SI}}=100\times [\frac{{\rm{s}}}{{\rm{s}}+{\rm{b}}}]$$

with s calculated as the weighted frequencies of Ba3-Ba4, Fu3-Fu4, Ca3-Ca5 and Om3-Om5 nematodes, and b representing the weighted frequencies of Ba2 and Fu2 nematodes57.

Statistical analyses

Soil microorganisms and invertebrates

Linear mixed-effects models (LMM) were used to analyse the effects of drought, NPK fertilization, season, and their interactions on invertebrate feeding activity, microbial activity, and microbial biomass using the R-package “nlme”58. The random intercept of the model was structured with plots nested within blocks, nested within year (year as a categorical factor). To account for repeated measurements within plots, we compared first-order autoregressive and compound symmetry covariance structures based on the Akaike information criterion (AIC). As differences between AIC values were lower than 2, the simplest covariance structure (i.e. compound symmetry) was used. Based on the importance of soil water content for microbial activity and biomass59, soil water content was added as an additional explanatory variable to the LMMs (Tables S3 and S4, Figs S5 and S6). As we were expecting a strong relation between aboveground plant biomass and microbial biomass60, additional LMMs were used to test the influence of plant biomass on microbial biomass (Table S5, Fig. S7). To evaluate model variation explained by fixed and random effects, marginal and conditional R2 were calculated using the “MuMIn” package61; marginal R2 represents model variation explained by fixed effects in the final model and conditional R2 represents model variation explained by both random and fixed effects.

Nematode community indices

LMMs were also used to assess the effects of drought, NPK fertilization, season (spring and summer 2017), and their interactions on nematode indices, i.e. Enrichment Index, Structure Index, Channel Index, and Maturity Index. A random intercept with plots nested within block was included in the models. We accounted for repeated measurements within plots by using a compound symmetry covariance structure, which fitted the data better than a first-order autoregressive covariance structure based on the AIC. To evaluate model variation explained by fixed and random effects, marginal and conditional R2 were calculated using the “MuMIn” package61.

Soil pH

We also used LMMs to assess the effects of drought, NPK fertilization, and their interaction on soil pH for all seasons in 2017. Additionally, we tested the influence of soil pH on invertebrate feeding activity. A random intercept with blocks nested within season was included in the models. We accounted for repeated measurements within plots by using a compound symmetry covariance structure, which fitted the data better than a first-order autoregressive covariance structure based on the AIC. To evaluate model variation explained by fixed and random effects, marginal and conditional R2 were calculated using the “MuMIn” package61.

Nematode density, richness, and trophic groups

Generalized mixed-effects models (GLMM) were used to assess the effects of drought, NPK fertilization, season (spring and summer 2017), and their interactions on nematode richness, total density (i.e. total number of individuals in the nematodes community) and the abundance of each trophic group (i.e. percentage of individuals in each trophic group). Nematode richness and total density of nematodes were modelled with Poisson distribution (link = “log”), while the trophic groups were modelled with Binomial distribution (link = “logit”). The random intercept of the model was structured with plots nested within blocks. To account for over-dispersion, an observation-level random effect was used in the model with omnivorous and predatory nematodes as a response variable.

Nematode functional guilds and cp groups

GLMMs were also used to assess the effects of drought, NPK fertilization, and their interactions on nematode functional guilds and cp-groups (Table S6) using Binomial distribution (link = “logit”). The random intercept of the model was structured with plots nested within blocks, nested within sampling (sampling as a categorical factor). GLMMs were performed using the “lme4” package99,100. With the methods applied in our study, however, we can only speculate about potential changes in the soil faunal community composition. This highlights the need for future research to detect which specific groups are responsible for bait perforation. This could be done, for instance, by exposing bait lamina strips with a labelled substrate under controlled laboratory conditions101,102. Building on that, the abundances of the most important groups of soil organisms could be monitored in the field, while being exposed to different global change drivers.

In contrast to the invertebrate feeding activity, microbial activity was not significantly affected by the interaction of the two global change drivers. Moreover, we could not detect any interactive effects on nematode indices or nematode groups. This illustrates the robustness of a large portion of the soil community to interactive global change effects, which might therefore be able to buffer prospective global change effects to a certain extent.

In conclusion, the main groups of soil organisms investigated in the present study responded differently to the individual and interacting effects of global change drivers. Soil invertebrate activity was strongly impaired by both global change drivers and their interaction, while microbial biomass benefited from enhanced nutrient availability, and microbial activity was surprisingly unaffected by all treatments. Despite the strong seasonal dynamics of temperate regions, these treatment effects remained constant across all seasons within two years. Notably, nematode indices pointed to changes in the state of the ecosystem, shifting towards simplified and more disturbed systems under drought and especially under fertilization that mostly facilitated opportunistic species. We could show that soil biota differ considerably in their sensitivity to global change drivers and in their seasonal dynamics – also highlighting the importance of integrating seasonal effects into experimental frameworks. This may lead to far-reaching alterations of crucial ecosystem processes, since decomposition and nutrient cycling are driven by the interdependent concurrence of soil microbial and faunal activities46. By covering a range of different taxonomic and trophic levels of soil organisms, we could therefore show that single as well interacting global change drivers induce complex changes in soil food webs and functions.