This study was performed at the Jena Experiment field site (51) on an Eutric Fluvisol (FAO-Unesco 1997) developed from up to 2 m thick loamy fluvial sediments alongside the Saale river in Jena, Thuringia, Germany (50°55’N, 11°35’E, 130 a.s.l.). The mean annual precipitation is 610 mm and mean annual temperature 9.9°C (57). The site was used as arable field for 40 years, before a conversion to grassland in 1960s, and was highly fertilized prior to the experiment setup (sowing of grassland communities) in 2002. The field site is split into 80 large plots of 20 x 20 m containing different plant community compositions varying in both species richness (1, 2, 4, 8, 16, and 60 plant species) and number of functional groups (1-4 groups: grasses, small herbs, tall herbs - summarized here as ‘herbs’ - and legumes) as described in Roscher, Schumacher et al. (51). The plots are distributed among four blocks to account for spatial edaphic variations (including soil texture and water holding capacity) which is related to the distance of the plots to the river. Each block contains the same number of plots at each plant species richness level, covering the range of functional groups (51). Overall, there are four replicates per species richness x functional group richness combination, with plant species in different communities chosen randomly out of a pool of 60 mesophilic grassland species. The sown grassland communities are maintained by three weeding campaigns per year in March, June, and October. There are two mowing events per year, in June and September.

The drought experiment was established in 2008 as a sub-experiment nested in the pre-existing plots. Prior to the second mowing in September, transparent rainout shelters (wood and PVC sheets, 2.6 x 3 m) were installed for 6 weeks every year to induce a prolonged summer drought period (50) over a span of 9 years (2008–2016). Of the two sheltered subplots – hereafter referred to as treatments, one received no water after installation (‘drought’) and one received collected rain water after precipitation events (‘control’), thereby controlling for non-drought roofing effects such as altered light and temperature (58). The roof shelters excluded 39.5 mm precipitation in 2009 and reduced summer precipitation by an average of 42% in 2008-2014 (8, 46). In July 2013, the field site was completely flooded during a natural flood event (occurring once in 200 years), and the drought treatments were continued thereafter.

To gain information on the lasting effect of 9-year history of drought vs non-drought on fungal communities, bulk soil samples were taken one year after the last summer drought simulation in August 2017 from all plots except the 60-species mixes. Around 50 g of bulk soil were taken by pooling 3-5 soil cores of 0-15 cm depths per subplot. The soil was sieved at 2 mm for homogenization and to remove any plant material. Soil samples were stored at -80°C until processing for next generation sequencing. As bulk soil was sampled one year after the last summer drought treatment, two 1 g subsamples were used for measuring gravimetrically the soil water content to ensure extant drought effect. For this, 2x 1g soil were dried at 104°C with Mettler Toledo and Kern DBS moisture analyzers.

To determine drought legacy effects on plant communities, plant aboveground biomass was determined by harvesting a 0.1 m² subplot at the end of August/begin of September 2017 by cutting plants with scissors at 3 cm above soil surface. The material was then sorted into target species, non-target weed, unidentifiable, and dead plant material and dried at 70°C for at least 48 h and weighed at weight constancy. This same procedure was also carried out throughout the 9 consecutive years of drought treatment, sampling aboveground biomass every year in May and August. Changes in aboveground plant biomass over the years were tested with ANOVA model:

Genomic DNA was extracted from bulk soil samples using DNeasy PowerSoil Kit (Qiagen). DNA quantity and purity was assessed using a NanoDrop 8000 spectrophotometer (Thermo Fisher Scientific). The SSU rRNA region of AMF was amplified in triplicates with a nested PCR using primer pairs Glomer1536 and WT0 [AATARTTGCAATGCTCTATCCCA/CGAGDWTCATTCAAATTTCTGCCC, Wubet, Weiß et al. (59), Morgan and Egerton-Warburton (60)] for the first and NS31 and AML2 [TTGGAGGGCAAGTCTGGTGCC/GAACCCAAACACTTTGGTTTCC, Simon, Lalonde et al. (61), Lee, Lee et al. (62) for the second PCR step following the protocol of Wahdan, Reitz et al. (63). To amplify the fungal ITS2 region, DNA was diluted to concentrations of 20 ng/µl and amplified in triplicates with primer pair ITS4 [TCCTCCGCTTATTGATATGC, Ihrmark, Bodeker et al. (64)] and fITS7 [GTGARTCATCGAATCTTTG, Gardes and Bruns et al. (65)] as described in Prada-Salcedo, Goldmann et al. (66).

Illumina sequencing libraries were prepared by purifying amplicons, barcoding and quality checking as described in Wahdan, Reitz et al. (63) and Prada-Salcedo, Goldmann et al. (66), respectively. Paired-end sequencing of 2 x 300 bp was performed at the Illumina MiSeq platform at the Department of Soil Ecology at Helmholtz Centre for Environmental Research – UFZ, Halle (Saale), Germany.

Raw SSU and ITS2 sequence data were filtered, trimmed, aggregated to amplicon sequence variants (ASV), and annotated with help of the dadasnake pipeline (67), which is based on DADA2 (68) using amongst other softwares cutadapt (69) and mothur (70). Within the dadasnake pipeline, AMF sequences were classified against SILVA 138 (71) and ITS2 sequences were classified against the eukaryote UNITE (version 8.2) database (72) using mothur’s implementation of the naïve Bayes taxonomic classifier (70). The final ASV tables were filtered for target taxa Glomeromycota and Fungi, respectively. This resulted in a total of 2,422,141 reads with on average 16,040 (SD +- 6,642) per sample of AMF. And for total fungi in 7,000,330 reads in total with an average of 46,360 (SD +- 6,202) per sample.

The taxonomic annotation of the AMF ASVs was not used further than to filter for Glomeromycota. Instead, AMF ASVs were aligned to virtual taxa (VTX) via MaarjAM’s BLASTn (73, 74). The read counts of ASVs assigned to the same VTX were summed up, condensing 4,865 ASVs to 114 VTX. To avoid uneven taxonomic resolution but include ASVs that are not yet represented in MaarjAM, ASVs that could not be assigned to a virtual taxon were extracted to construct a maximum likelihood phylogenetic tree based on a general time-reversible, discrete gamma (GTR+G) model using MAFFT (75) and raxML (76). Custom virtual taxa (VTC) with cophenetic distances below 0.03 were assigned. Their read counts were merged, resulting in a total of 211 virtual taxa (VTX plus VTC).

Obtained VT/ASV abundance and taxonomy results were further analysed in R version 4.0.4 (77). Abundances were sum-normalised and VTs/ASVs that were only present in up to 3 samples and with only one read were removed.

Data was rarefied to the lowest number of reads per sample (7,350 for AMF/27,000 for ITS) to calculate observed VT/ASV richness with phyloseq v 1.34.0 (78). Relative abundances were generated with transform_sample_counts calculating sample-wise fractional abundances in phyloseq. Jensen-Shannon divergence (JSD) which is effective at capturing compositional changes and commonly used in microbial ecology (79, 80), was calculated with vegan v 2.5-7 (81) and visualized by NMDS. Environmental factors were fitted onto the ordination with envfit and factors with significant correlations plotted as vectors onto the ordinations. Vegan was also used for PERMANOVA models (adonis2) and analysis of multivariate group dispersion (betadisper). Graphics were created with ggplot2 v 3.3.3 (82).

We used multiple models to test for different interactions of fungal communities with their environment. Multivariate effects of experimental factors on JSD matrices were assessed with the following three models:

1. To analyse whether the experimental factors drought and plant diversity have an interactive effect on fungal communities (stratified by plot):

2. To test whether natural variation of soil water content across the field site adds to the drought effects (stratified by plot):

and 3. to gain a better understanding of the differences in plant-fungal-interactions between drought and control conditions, a model of effects of plant community composition within the drought/control treatment plots was tested for each treatment separately (stratified by block):

Similar models were used for ANOVA for α-diversity, soil moisture and plant biomass, with the exception of plot/block being added as random terms.

We calculated matrix dissimilarities of Drought and Roofed Control VT/ASV abundances from similarity matrices (1 – JSD matrix) based on the modified RV coefficient using package iTOP (83, 84). We analysed differential abundances of ASV/VT in drought vs control subplots with the function ANCOM (85).

To test whether the impact of the repeated summer drought treatment was extant in the drought manipulated plots, we analysed soil water content of the bulk soil samples used for DNA extractions. Soil water content was significantly lower (t = -3.16, p< 0.05) in drought-treatment plots than in roofed controls. Soil water content was affected by the repeated drought treatments, but also by the plot position on the field site and the thereby existing variation in soil properties ( Figure S1 ). Separately analyzing drought from control plots, we found significant effects of plant species richness and block on soil water content in both drought and control plots ( Table S2 ). The effect size of these diversity and spatial gradients was reduced under drought.

Aboveground biomass of target plant species was determined for 2008-2016 every year in May and August. Tested against the sampling date, drought treatment and plant diversity we found plant biomass differed with sampling date (F = 17.61, p< 0.001) and plant diversity (F = 373.94, p < 0.001). However, testing biomass data of May and August separately erased the significance of the sampling date, while plant diversity effects remained, showing that the sampling date caused seasonal fluctuations instead of changes over the years. In May, plant biomass over the years was additionally affected by the interaction of drought treatment and plant diversity (F = 4.23, p = 0.04). Plant biomass significantly increased from August 2016 (last drought treatment) to August 2017 (F = 155.95, p< 0.001) influenced by plant diversity (F = 40.79, p< 0.001). This increase in biomass after the last drought treatment was consistent for all plant functional groups: legumes (F = 8.59, p > 0.01), grasses (F = 10.27, p< 0.01) and herbs (F = 10.57, p< 0.01), however, biomass of the plant functional groups individually was not affected by the drought treatment or plant diversity ( Figure S5D ).

At the time of soil sampling in August 2017, total plant biomass per plot was slightly lower in former drought-exposed than in control plots, but not significantly (t = -0.64, p = 0.52). The ANOVA model for total plant biomass showed no effect of the drought treatment (F = 0.41, p = 0.53), but instead of the specific soil water content per plot (F = 20.72, p< 0.001; Table S2 ). Considering plant community composition as a factor for plant biomass, we found plant species richness to positively affect total plant biomass in both drought and control (F = 8.19, p< 0.01), with the effect size increasing under drought (F = 19.41, p< 0.001). Along the plant diversity gradient, drought did not significantly change plant biomass (neither in species-poor nor species-rich communities). The absence of legumes significantly lowered plant biomass, overall (t = -4.93, p< 0.001), and the effect size of legumes on total plant biomass increased under drought (drought: F = 6.12, p = 0.02, control; F = 5.98, p = 0.02). The biomass of legumes also slightly increased under drought, as opposed to grasses and herbs that declined in biomass. The differences in plant biomass were however not significant ( Table S1 ).

To test whether the drought and plant species and functional richness treatments have lasting effects on the composition of total fungal and AMF communities in the grassland plots, we calculated distance matrices based on Jensen-Shannon divergence and conducted PERMANOVAs. The first model for drought and plant diversity interaction showed significant effects with low effect size for both the drought treatment (R2 = 0.01, p< 0.001, Figure 1C ) and plant diversity (R2 = 0.07, p< 0.001) on AMF communities and no interaction of drought and plant diversity (R2 = 0.01, p = 0.06). The variation of soil water content within the drought treatment had no significant effect on AMF communities For the AMF community composition, the overall model for effects of drought revealed evidence of a small treatment effect (R2 = 0.01, p< 0.001) and no significance of soil water content (R2 = 0.06, p = 0.09, Figure 1E ). To assess correlation of abundance matrices of drought and control treatments, we calculated the correlation coefficient for similarity matrices (modified RV). AMF drought and control matrices showed a significant correlation coefficient (RV = 0.43, p< 0.001). Within the drought treatment plots, AMF community composition was affected by plant species richness and presence of grasses and legumes, which explained a total of 17% of variation in the beta-diversity ( Figure 1G ). Presence of herbs and number of functional groups present were not significant determining factors of AMF communities under drought. In control plots however, all plant composition factors except presence of herbs played a role in AMF community structuring: plant species richness, presence of grasses and legumes plus the number of functional groups present amounted to 21% explained variation. The low effect size of all these factors is apparent in the non-metric multidimensional scaling of the JSD matrix, as neither treatment nor plant community cause strict clustering of communities ( Figure 1A ). A test for homogeneity of dispersion within AMF communities was not significant for treatments (F = 1.29, p = 0.25) nor for plant diversity (F = 2.27, p = 0.06) suggesting homogeneous dispersion of groups.

For the beta-diversity of the total fungal communities ( Figure 1B ), the first model on drought and plant diversity interactions revealed significant effects with low effects size for the drought treatment (R2 = 0.01, p< 0.001, Figure 1D ) and plant diversity (R2 = 0.07, p< 0.001), however these two factors did not interact (R2 = 0.00, p = 0.63). The variation in soil water content tested with PERMANOVA model 2 does not add additional variation to total fungal communities (R2 = 0.07, p = 0.20, Figure 1F ). The coefficient of correlation between drought and control similarity matrices was higher than in the case of AMF with RV = 0.74 (p< 0.001). Within the drought exposed plots, plant species richness, presence of grasses and legumes, but not herbs, shaped total fungal communities with a total explanatory value of 17.6%. Total fungal communities under drought were not affected by the number of functional groups present (R2 = 0.07, p = 0.11). The influence of these plant community factors was similar in significance and effect size in control plots ( Figure 1H ), however in control plots herbs also determined fungal community structures (R2 = 0.02, p = 0.02). Tests for homogeneity of dispersion within ITS communities were not significant for the drought treatment (F = 0.22, p = 0.64), but significant for plant species richness (F = 3.34, p = 0.01). An adhoc TukeyHSD further revealed different dispersion between groups of 4 and 1 plant species (padj = 0.05) and 16 and 1 plant species (padj = 0.05), suggesting that parts of the observed differences were due to differences in dispersion rather than between the treatments.

Differential abundances were analysed with ANCOM (Figure 2). Among the AMF were only the two taxa VTX00354 (Diversispora Torrecillas 12b) and VTC00015 (Paraglomus sp) significantly differentially abundant – both higher abundant under drought with log fold changes of 1.15 (VTC00015) and 1.21 (VTX00354). Within the total fungal communities, we found three taxa that were significantly higher abundant under drought conditions: Penicillium jenensii with a log fold change of 1.85, an unidentified Penicillium sp. (log fold change = 0.33) and an unidentified Ascomycot sp. (log fold change = 1.02).

In brief, AMF and total fungal communities were shaped by plant community composition and the drought treatment caused lasting shifts in community structures. However, drought and plant community explained only small proportions of the AMF and total fungi community structures.

Given the observed effect on community composition, we also tested whether the α-diversity of AMF and total fungi was affected by the repeated drought treatment. Paired t-tests of VT/ASV richness in drought and control subplots showed no significant differences for neither AMF (t = 0.39, p-value = 0.69) nor total fungal (t = 1.70, p = 0.09) communities. Testing α-diversities against model 1, AMF VT richness was neither significantly altered by drought (F = 0.13, p = 0.72) nor plant diversity (F = 0.12, p =0.73) nor their interaction (F = 2.23, p = 0.14). Total Fungi ASV richness was not altered by drought (F = 1.29, p = 0.26) or an interaction of drought and plant diversity (F= 0.02, p = 0.88), but showed marginal significance for a plant diversity effect (F = 4.08, p = 0.05).

When applying ANOVA with the drought treatment model to α-diversity, neither drought treatment as a factor nor specific SWC were influencing α-diversity of AMF or total fungi ( Figure 3 and Table S3 ). Regarding the influence of plant richness and presence of functional groups on fungal α-diversity, plant species richness (R2 = 0.13, p = 0.03) and presence of herbs (R2 = 0.05, p = 0.03) increased α-diversity of AMF in control plots, but no plant functional group was significant for AMF α-diversity under drought ( Table S3 ). In contrast, the presence of herbs (R2 = 0.06, p = 0.03) and legumes (R2 = 0.05, p = 0.04) significantly increased total fungal communities under drought, but not in control plots. The fungal α-diversities in more diverse plant communities were not affected by the drought treatment.

In short, drought had no major impact on fungal α-diversity. Plant functional groups played a role under drought for total fungal communities, but not for AMF. AMF α-diversity however was affected by plant communities in control plots.

The 9-year history of drought treatment effectively changed the soil water content, which appeared still significantly lower in plots that were exposed to repeated drought even one year after the last treatment. Previously, Vogel, Fester et al. (58) found soil moisture to not differ significantly between the treatments in the same experiment after one year, but found a response of soil moisture to the drought treatment over time, as the difference was significant in the third year of drought treatment. This is congruent with the soil water content being still divergent one year after the end of treatment, despite the field site flooding in 2013 (86). The reduction amounted to only about 2% less soil water in the drought treatment compared to control plots. SWC additionally varied across the field site, coinciding with edaphic variation of the field site (87) which was accounted for by always including the field block as error term. Overall this shows that even with a moderate, but repeated, drought treatment, changes in ecosystems become visible, and we need even more experimental approaches of realistic climate scenarios as opposed to climate extremes (88) to understand consequences on multiple ecosystems levels.

Arbuscular mycorrhiza have repeatedly been shown to enhance drought tolerance of their plant partners (89). Previous studies have shown that plants rely on mycorrhizal symbiosis under drought stress, but that AMF colonisation decreases under drought (90, 91). Soil moisture regulates AMF community assembly, with drought causing shifts in relative abundances including the absence of certain AMF, which results in a reduced α-diversity (92). Therefore, we expected to see changes in the α-diversity of AMF in response to repeated drought. Our treatment with nine years of prolonged summer droughts, however, had no significant impact on AMF α-diversity (VT richness). Neither the repeated drought treatment nor the resultant SWC had significant effects on AMF α-diversity, therefore drought in our experiment did not cause a recruitment of more or less species. Analyses of the root and rhizosphere mycorrhiza may yield stronger α-diversity effects because they represent the place of AMF-plant interaction. In this regard, we also have to take into account that the flooding of the field mid-experiment might have been a mode of transport for propagules.

Despite the α-diversity of AMF remaining similar under drought across the diversity treatments, the community composition did change. This effect was apparent regardless of the high variation of SWC along the field and across the plant diversity gradient. However, the variation explained by drought was limited, which is in line with AMF having been shown able to endure drought (36). Two AMF VT were found to have a higher relative abundance under drought. The first one was identified as Diversispora, whose mycorrhizal infection is inhibited by drought (93). But Diversispora were found to be present predominantly outside rather than inside plant roots (94, 95) exhibiting low infection rates of roots but proofing effective for plant productivity (96). These edaphophilic AMF show high biomass allocation to extraradical hyphae and increase plant nutrient uptake, but are usually characterized as sensitive to drought (97). With moderate drought as we had in our experiment, this drought sensitivity of Diversispora might however have been overruled by their interaction with plants. The second AMF genus with increased abundance under drought was Paraglomus. This genus is rhizophilic with high abundance in grass roots (97) and was previously reported as having a positive influence on drought resistance of plants (98, 99). A higher relative abundance of these two genera under drought suggests distinct responses of AMF species in relation to their host preference and their drought response. However, this is in terms of relative not quantitative abundance and might also be results of these genera being persistent while others decrease.

Summing up the effects of repeated summer drought treatment on AMF communities, we found minimal drought-induced changes on AMF α- and β-diversity, but the differential abundances of an edaphophilic and rhizophilic AMF indicate shifts in the interaction of AMF with plant partners. A reason for the relatively small drought effects may be the sampling of bulk instead of rhizosphere soil. Even though bulk soil in grasslands is still in vicinity of and highly shaped by plant roots (100), we cannot be sure that our results would be identical in the rhizo- or endosphere. Rhizosphere soil is subject to stronger changes through plant nutrients and biomass and other microbes like bacteria (101) and even in the confines of pot-experiments has been reported to respond stronger to drought stress than bulk soil (102). Nevertheless, resistance of bulk soil communities is important for potential feedback on plant communities because they represent the general pool of available interaction partners.

Recurring drought events affect ecosystems negatively, however plant species richness proved to be an important factor of resistance and alleviation (3, 4, 8, 46, 103). As Wagg, O’Brien et al. (8) reported, the plots with high plant diversity maintained long-term productivity under drought in our experiment. Similarly, in the year after the last drought treatment, we found aboveground plant biomass to be lower after repeated drought with the reduction of productivity tending to be higher in less diverse plant communities. The influence of plant diversity as well as presence of legumes on biomass increased under drought. Noteworthy, legumes were the only plant functional group with an increase in biomass after drought exposure.

Throughout the repeated drought treatments plant biomass was overall not greatly reduced by drought. The recurring drought however induced seasonal shifts: productivity in May was more susceptible to drought and drought-diversity interaction as already shown for the first 5 years of recurring drought by Wagg, O’Brien (8). It seems, especially highly diverse plant communities had stumped productivity, however, all drought exposed plant communities could recover quickly after the last drought exposure. This is in line with findings of Chen, Vogel (104) who collected seeds from this same experiment, and exposed the plants of drought or ambient legacy once again to drought and found drought-legacy plants to recover much quicker after droughts than ambient-legacy plants.

By sampling a year after the last drought treatment, we tested for a drought legacy effect rather than a drought response. Plants have been shown to adapt to drought stress and even establish a drought memory with physiological and morphological alterations (105). Changes in precipitation patterns can reduce net primary production and affect e.g. soil respiration and nitrogen mineralization (106). Microbial processes respond quickly to changes in soil moisture, but can revert back just as quickly (105). Contingently, the shifts in microbial communities we found are a result of the plants’ drought memory.

With plants shaping belowground processes, we expected to see a positive effect of plant species richness on AMF community compositions in both control and drought plots and a buffering effect against the drought effect by higher plant diversity. Correlations of the distances between community compositions as well as the PERMANOVA models showed that total fungi and AMF were affected by plant species richness and community composition, independent of drought events. Likewise, our results suggest that a higher plant diversity does not buffer negative drought effects on soil fungi diversity. This lack of effect might reflect that the drought effect on soil moisture was weak across the plant diversity gradient. In addition, while soil under richer plant communities had a higher water content in control plots, this relationship decreased under drought, indicating that the long-term buffering effect of diverse plant communities on the soil water content was also weak. Finally, plant species richness was an important factor for β-diversity as well in controls as under drought stress. Our results are in accordance with data from the beginning of the experiment (50) showing that after one year of drought, higher plant diversity did not buffer against belowground drought effects, such as lower litter decomposition rates and microbial respiration. These findings suggest that plants in high diversity settings maintain their aboveground rather than belowground productivity under short-term adverse conditions. In our study, plant biomass was not significantly reduced under drought at any plant diversity level. However, increasing effect size of plant species richness on biomass under drought indicate stronger differences between the diversity levels. The described maintenance of plant productivity could be explained by short-term variation in biomass production, which increased in more diverse communities due to compensatory recovery from drought (8). Considering this drought-induced short-term variation, buffering effects of diversity might become more visible with seasonal variations, and Wagg, O’Brien et al. (8) found spring growth to compensate for summer drought induced productivity losses but only at high plant diversity. Our data comes from just a single time point in the driest month of the year, which may have caused us to miss temporary changes. Thus, we would expect those short-term changes to occur predominantly in the rhizo- or endosphere, while the bulk soil microbiome reacts less to droughts (102, 107).

Correlations of AMF abundance and plant species richness have been previously shown, for example, AMF abundance decreasing in the presence of certain grass species and increasing in the presence of legumes (55). Legumes have been proven to facilitate carbon allocation to fungi under moderate drought stress (108). Our results showed that the number of different functional groups is only relevant for α-diversity of AMF in drought but not control plots, while it shaped β-diversity only in controls. However, presence of grasses and legumes promote AMF communities in both control and drought with the influence of legumes even increasing under drought. The reduced effect of grasses under drought on AMF communities might stem from grasses raising the soil water content in the short term (50), an effect which did not persist to the end of the experiment according to our soil water content measurement. Kivlin, Mann et al. (54) found plant identity to be a major factor in shaping fungal communities despite environmental gradients, but diverse climate changes can interrupt plant-microbe interactions (109). Drought in particular affects water uptake and nutrient concentration, such as C:N ratio, leading to increased plant-plant competition (110). Some functional plant groups benefit more from AMF, as Chagnon, Bradley et al. (111) suggest: limitation of carbon allocation to root exudates caused by stress negatively affects plant-soil interactions, but slow growing plants invest more in AMF and receive more long-term benefits. Additionally, AMF have varying effects on productivity of different plant functional groups. Legumes seem to benefit in terms of whole plant growth while mycorrhization for herbs increases reproductive growth (112). This may explain the stronger effect of the presence of legumes on AMF communities. Legumes use more water, but have a higher water use efficiency than e.g. grasses (113) and together with their facilitated carbon allocation towards fungi under drought become a more desirable interaction partner for fungi. Microbes vice versa intensifying plant-plant competition under drought (110) could further explain the increase in legume biomass and legume-fungal interaction confirming long-term benefits of AMF symbiosis. AMF have host preferences (43) which might increase under stress conditions in relation to available transferable nutrients. Interestingly, AMF α-diversity was higher in monocultures even though AMF host-preferences would suggest higher AMF diversity with increasing plant diversity (114, 115). However, monocultures seem to benefit from AMF colonization even under drought, while plant mixtures might encounter disadvantages from demanding AMF species (116).

To test whether AMF as biotrophic interaction partners of plants are specific in their response to drought, we compared their response to that of other fungi. Glomeromycota made up 3.5% of abundance in control plots within the ITS2 dataset and increased to 4.1% of abundance under drought. This increase was expected: Even though fungal communities in general are rather resistant to drought (30), they have been shown to shift towards a lowered abundance in Agaricomycetes and other fungal groups and a higher abundance in Glomeromycota (117). However, the low proportion of Glomeromycota is due to the choice of the general fungal ITS2 primers, which bias amplification against the Glomeromycota (118, 119). In addition, most ITS2-reads of Glomeromycota could not be identified at the genus level. Therefore, the results discussed in the previous sections stem from AMF-specific amplifications and the rest of the community was assessed independently.

The α-diversity of total fungal communities did not change in response to drought, as for the AMF. But the explanatory value of legumes for general fungi increased under drought, indicating that the increasing biomass of legumes under drought plays a role for general fungal diversity and maybe a shift towards more AMF species. Sweeney, de Vries et al. (55) showed not only that AMF have a preference for legumes driven by root traits, but also that higher abundance of AMF in roots leads to a reduced abundance of fungal pathotrophs.

Both AMF and the total fungi β-diversities did minimally change under repeated drought, but they differed in their response to plant communities under adverse conditions. In the control plots we found AMF to be more responsive to plant species richness than the total fungi communities. The effect of legumes and herbs was similar for both fungal communities, but total fungi are additionally shaped by grasses, while for AMF communities the plant functional diversity plays a role. Total fungi communities under ambient and stress conditions were consistently affected by plant diversity and functional composition. This is in line with several studies showing that the fungal community is driven by plant functional identity and root traits with e.g. saprotrophic fungi being less abundant in grass roots (55) and plant functional diversity being a major driver for fungal richness and community structures even under stress conditions like drought (120, 121).

While we found two significantly higher abundant AMF VT under drought, the total fungi community did not detect any differential AMF at genus level. Here we found Penicillium to be more abundant under drought and no genera significantly decreasing under drought stress. One of the two Penicillium, P. jensenii, is often found in locations with higher moisture (122), but as saprophytic and phosphate-solubilizing fungi they might thrive under drought as more dead plant matter becomes available (123).

Comparing the responses of AMF and total fungi in our experiment, AMF communities displayed stronger shifts under drought than the total fungal communities. The drought-induced shifts were towards a slightly higher richness and abundance of AMF. Juxtaposing AMF communities and total fungi communities emphasizes even more the importance of plant functional identity as driver of fungal soil communities and plant-fungal interactions.

In summary, the effect of nine consecutive years of prolonged summer drought events on fungal communities was minimal across the plant diversity gradient. This means soil-borne fungi, including AMF, were similarly resilient to drought at lower and higher plant species richness. Fungal α-diversity was not negatively affected by drought, and β-diversity of AMF and total fungi communities only slightly shifted. The fungal communities therefore did not seem to be protected from drought by plant diversity, in contrast to plant productivity, which was more strongly affected by plant species richness under drought. All these findings suggest that plant diversity itself does not shape fungal community structures under drought stress. We also found that the influence of plant functional diversity on AMF communities decreased under drought, while it increased for general fungi. However, the effect of the presence of legumes increased under drought. While overall plant biomass productivity did not change greatly in response to repeated yet moderate drought exposure, shifts we do find in fungal communities indicate indirect drought effects caused by changes in plant-fungal interactions.