Soil samples were collected from 10 sites in the Trentino province (Additional file 1: Figure S1 and Additional file 14: Table S1a). Each site was characterized by sampling the vineyard (V) and the grasslands at 8 (P1) and 16 m (P2) from the grapevine row (see “Methods” section). For each triplet, 6 samples were collected, for a total number of 180 samples. After discarding 3 samples due to the low quantity of extracted DNA, a total of 5,705,432 amplicon sequences from the V4 region of 16S rRNA gene and 7,350,959 sequences from the ITS region were obtained from the remaining 177 samples. The 16S and ITS sequences were clustered into 21,113 and 12,542 operational taxonomic units (OTUs, 97% identity), respectively. After dropping one 16S sample due to the low number of reads, 16S and ITS samples were evenly rarefied to 16,000 and 22,000 reads per sample, respectively. After rarefaction, the dataset was composed of 19,584 bacterial and 12,101 fungal OTUs.
Microbiota composition is conditioned by cultivation for both bacterial and fungal communities
The dominant bacterial Phyla were Acidobacteria (22.7%), Proteobacteria (18.8%), and Actinobacteria (16.5%), while on average, 14.1% of the reads could not be classified. At the family level (Fig. 1a, c), the fraction of bacterial OTUs that could not be classified grew to 34%, while for those that were classified the dominant family was Gp6 (13%), followed by Nitrosospheraceae and Planctomycetaceae (9% and 5%, respectively). For fungi, the dominant Phyla were Ascomycota (51.8%), Zygomycota (20.1%), and Basidiomycota (11.2%), while 12.1% was constituted by unclassified OTUs. At the family level (Fig. 1b, d), the dominant taxa were Mortierellaceae (17.4%) followed by Nectriaceae (8.8%) and a family of unidentified Ascomycota (5.2%). The fraction of fungal OTUs not classified at the family level was 25%.
To identify taxa that were significantly impacted by cultivation, we modeled the read counts using generalized linear models [39] (GLMs) taking baseline differences between the sites into account. For this analysis, samples from both types of permanent grassland (both P1 and P2) were considered together. For bacteria, we found 336 OTUs that were significantly differentially abundant (p < 0.01, FDR corrected) between samples from vineyard soils and from permanent grasslands from the same site. Of these, 224 were higher in vineyard samples and 112 in grasslands (Additional file 15: Table S2). The 10 most significant ones included taxa from the families Gp4, GP6, Hyphomicrobiaceae, and Planctomycetaceae (Additional file 4: Figure S4). For fungi, we found 57 taxa significantly (p < 0.01, FDR corrected) more abundant in the vineyard than permanent grassland soil from the same site, and 37 taxa significantly less abundant in vineyard samples (Additional file 16: Table S3). Amongst these, the most significant were taxa from the family Amphisphaeriaceae, that in vineyard samples accounted for up to 20% of the fungal microbiota (Additional file 5: Figure S5). Other differently distributed taxa included unidentified Pleosporales, unidentified Ascomycota, Hyaloscyphaceae, and Sordariomycetes.
Bacterial communities have a core of conserved species that is shaped by both location and land use
The distribution of OTUs across samples (Fig. 2a) showed that the largest fraction of OTUs was specific to a small number of samples while only a small set of OTUs was ubiquitous. Specifically, out of 19,584 OTUs, only 162 were present in all samples, 484 were present in at least 80% of the samples, and 961 in at least 50% of the samples.
Despite the relatively low number, the core OTUs accounted on average for 48 ± 3.8% of the bacterial reads of each sample. Using a more relaxed definition of the core OTUs, i.e., defining as “core” those OTUs that are present in more than 95% of the samples, core OTUs accounted on average for 64 ± 3% of each microbiome (Fig. 2b). For comparison, the specific OTUs (i.e., those present in less than 5% of the samples) that are the vast majority of the bacterial OTUs, only accounted for an average of 1.5 ± 0.5% of each bacterial microbiome. Acidobacteria, a minority component of the specific microbiome, was the dominant component of the core microbiome. Other frequent components of the core microbiome included Actinobacteria and Proteobacteria, while Bacteroidetes and Firmicutes were almost absent.
We next asked if the size of the core microbiome was influenced by the cultivation type or geographical origin of the samples. We found that the core microbiome of P1, P2, and V samples included 250, 256, and 240 OTUs, respectively, while the site-specific core microbiome varied between 372 and 638 OTUs (Additional file 17: Table S4). In both cases, the core microbiome was significantly larger than expected by random sampling (Wilcoxon rank-sum test, p values 1.16 × 10−6 for site and 7.8 × 10− 3 for cultivation type, respectively), suggesting that both these factors favor the colonization of soil by a defined set of bacterial species.
In general, in any given dataset, the number of core OTUs (core microbiome) decreases with the number of samples, since each newly added sample might miss OTUs that were core in the reduced dataset. Therefore, it is possible to estimate the true size of the core microbiome, i.e., the number of OTUs that are always present in this kind of soils independently of the sampling size by extrapolating from a random subsampling. The results are shown in Additional file 6: Figure S6a, where we plot the number of core OTUs as a function of the number of samples. The curve can be fitted by a power-law decay converging to a plateau of 117 ± 3.5 OTUs that is the estimated size of the core bacterial microbiome of these soils.
The core of conserved fungal species is small and determined by location, but not land use
Compared to bacteria and archaea, the fungal component of the soil microbiome (the mycobiome) was more variable across the different sampling sites. Indeed, out of 12,101 total OTUs, only 5 were present in all samples (Fig. 2c). Core OTUs accounted on average for 15 ± 5% of each sample (35 ± 8% for OTUs present in more than 95% of the samples, Fig. 2d), while the specific OTUs (i.e., those present in less than 5% of the samples) represented an average of 4.8 ± 0.3% of each mycobiome (Fig. 2d). From a taxonomic point of view, the core mycobiome was dominated by Ascomycota, Zygomycota, and a small fraction of Basidiomycota. Using the same subsampling strategy outlined above, the size of the core mycobiome as a function of the number of samples was again well described by a power-law (Additional file 6: Figure S6b) that in this case converged to a value close to zero (4.1 ± 0.4 OTUs). Differently to what found for bacteria, the core of conserved fungal species in the three cultivation types (17, 10, and 13 OTUs for V, P1, and P2 samples, respectively) was not significantly larger than expected by random sampling (p value 0.58). On the contrary, the different sites had a site-specific core mycobiome (Additional file 17: Table S4) significantly larger than expected by random sampling (p value 3.2 × 10−4). These results suggest that geographical factors dominate the composition of the soil mycobiota that show a higher level of variability compared to bacteria, and that cultivation is not able to select a defined set of fungal species across different locations overcoming the differences due to geographical factors.
Bacterial and fungal richness vary with site and cultivation, and are correlated in a site-specific manner
The microbiome richness, or α-diversity, varied widely across locations, sites, and cultivation type (Additional file 18: Table S5 and Additional file 19: Table S6) for both bacteria and fungi (Fig. 3a, b respectively). We first tested the differences across the locations (Additional file 18: Table S5) and sampling sites (Additional file 19: Table S6) finding that in many cases the differences were significant. We then tested whether the microbiota richness was significantly different between the three cultures within the same sampling site. Despite the fact that samples from the vineyards had often a bacterial α-diversity significantly different from surrounding permanent grassland, we could not highlight an unambiguous effect of cultivation across all sites (Additional file 20: Table S7). For bacteria, in no case the P1 and P2 samples had significantly different richness (exceeding the 0.05 significance level), while in three cases the V samples have a significantly higher, and in one lower richness than the corresponding P1 samples (Additional file 20: Table S7). For fungi, in two cases, the α-diversity of the P2 samples were significantly different from the P1 samples, while in three cases, the V samples had a lower α-diversity than the corresponding P1 samples, and in one case higher (Additional file 20: Table S7).
To highlight the possible effects of competition between bacteria and fungi, we tested by linear modeling whether there was a correlation between the bacterial and fungal α-diversities in the same samples. We found that the correlation across all samples was negligible (adjusted R2 = -0.006, p value = 0.86, Additional file 21: Table S8). However, correlating the diversity indexes within each site the correlation was significant (adjusted R2 = 0.22, p value = 2.22*10−6, Additional file 22: Table S9). In most sites, the fungal α-diversity was negatively correlated with the bacterial α-diversity, with the exception of PT05 and PT12 both from the same area (Ala), where we found a positive correlation between bacterial and fungal α-diversity (R2 = 0.51, slope 0.91, p value = 0.0016, Additional file 23: Table S10 and R2 = 0.48, slope 0.64, p value = 0.0008, Additional file 24: Table S11, respectively). For the other sites, using a linear mixed-effect model where the intercept was treated as random variable, we obtained an estimated value of − 0.59 (p value = 0.002, Additional file 25: Table S12) for the slope of the correlation between the fungal and bacterial α-diversity.
Chemical characteristics of the soil partially explained the variability in richness
To identify soil features that have an impact on α-diversity, we built a random forest model including the texture (percentage of sand, silt, and clay) and chemical characteristics of the soil (absolute quantity of CaCO3, Cu, Zn, Pb, Cd, soil organic matter, carbon-to-nitrogen ratio-C/N -, and pH). The model was able to account for 58.25% of the α-diversity variability for bacteria and archaea. The more relevant characteristics were the absolute quantity of CaCO3, the percentage of sand, and the percentage of silt, followed by the absolute quantities of Zn and Cu, and by pH. Differently from what recently found in global surveys [7, 12, 45], the effect of pH was moderate, probably due to the relatively small range of pH values sampled in the present study. By linear modeling, we found that the bacterial microbiome richness was significantly positively correlated (Additional file 7: Figure S7 and Additional file 26: Table S13) with CaCO3 and silt and negatively with sand, Cu, and Zn. The correlation with pH was not statistically significant.
For fungi, a random forest model explained only 19.89% of the variability of the α-diversity. The most relevant factor was the concentration of Cu, followed by the concentration of silt, Zn, total N, organic matter, and sand. A linear model showed that the concentration of Cu and Zn were negatively correlated to the α-diversity of the fungal microbiota (Additional file 8: Figure S8 and Additional file 27: Table S14). The correlations with silt, total N, organic matter, and sand were not statistically significant.
The characteristics of the soil had a large impact on the structure of the microbiota
We performed a maximal information coefficient analysis [42] at all taxonomic ranks to identify taxa that were sensitive to specific characteristics of the soil (Additional file 28: Table S15 and Additional file 29: Table S16, respectively). The characteristics of the soil correlated differently with different bacterial and fungal clades (Figs. 4, 5 and Additional file 9: Figure S9 and Additional file 10: Figure S10). The ratio between the relative abundances of bacteria and archaea was associated with all the measured quantities, with the exception of the C/N ratio and the concentration of Cd. In some cases (Additional file 11: Figure S11 and Additional file 28: Table S15), these associations correspond to non-zero values of Spearman’s rho (ρ = 0.54, MICe = 0.38 for the concentration of Cu and ρ = − 0.44, MICe = 0.37 for the relative abundance of sand, respectively), while in other cases, the Spearman’s rho was close to 0 (ρ = − 0.09, MICe = 0.22 for the relative abundance of clay). In general, the strongest associations (positive and negative, respectively) in 16S data are between the relative abundances of two bacterial families, namely, the family Verrucomicrobiaceae of the order Verrucomicrobiales [46] (ρ = 0.72, MICe = 0.51) that was positively correlated with the amount of silt, and the family Phyllobacteriaceae [47] of the order Rhizobiales (ρ = − 0.81, MICe = 0.58), that was negatively correlated with the concentration of Cu (Additional file 12: Figure S12 a and b, respectively).
For fungi, several clades were strongly correlated to the characteristics of the soil (Fig. 5 and Additional file 10: Figure S10). The strongest associations were between one OTU from the family Herpotrichiellaceae, that was negatively correlated with the relative amount of silt (ρ = − 0.80, MICe = 0.72) and positively with the relative amount of sand (ρ = 0.71, MICe = 0.55).
The composition of the bacterial microbiome correlates with geography, while fungal communities are dominated by cultivation type
To explore if differences in microbiome structure and composition correlate with sampling location and soil type, we computed the between-sample diversity (−diversity) using Bray-Curtis distance (Fig. 6). For bacteria, samples from the same location and type of soil generally clustered together (Fig. 6a). In addition, samples from both types of permanent grassland soil were closely related, with only one case in which a PERMANOVA test indicated a clear distinction (PT16, Additional file 29: Table S16) and two in which the test was marginally significant (p value = 0.01, PT05 and PT17). Differently, in all locations, the vineyard soils were clearly distinct from the corresponding permanent grassland soils (Additional file 29: Table S16). However, in most cases, samples from the same location formed well-defined groups and the distances between permanent grassland and vineyard samples from the same site were smaller than the distances between samples from different sites (Additional file 13: Figure S13).
For fungi (Fig. 6b), samples from permanent grassland soils were closely related (with two exceptions, namely, PT01 and PT16, and one, PT17, for which the PERMANOVA test was marginally significant, i.e., p = 0.01, Additional file 29: Table S16), and were distinct from corresponding vineyard samples. However, in contrast with what was found for bacteria, a large fraction of the samples from the vineyards formed in a large group (Fig. 6b), samples from the same site were in most cases similar to the distances between samples from different sites (Additional file 13: Figure S13). These results suggest that soil usage has a stronger effect on the fungal component of the soil microbiota than on bacteria.