Volume 209, Issue 3 p. 1196-1207
Full paper
Free Access

Contrasting microbial biogeographical patterns between anthropogenic subalpine grasslands and natural alpine grasslands

Roberto A. Geremia

Corresponding Author

Roberto A. Geremia

Lab Ecol Alpine LECA, Université Grenoble Alpes, F-38000 Grenoble, France

CNRS, Lab Ecol Alpine LECA, F-38000 Grenoble, France

Author for correspondence:

Roberto A. Geremia

Tel: + 33 0 676 259 391

Email: [email protected]

Search for more papers by this author
Mihai Pușcaș

Mihai Pușcaș

‘A. Borza’ Botanical Garden, Babeș-Bolyai University, 42 Republicii Street, 400015 Cluj-Napoca, Romania

Search for more papers by this author
Lucie Zinger

Lucie Zinger

Lab Ecol Alpine LECA, Université Grenoble Alpes, F-38000 Grenoble, France

CNRS, Lab Ecol Alpine LECA, F-38000 Grenoble, France

CNRS, ENFA, UMR 5174 EDB, Université Toulouse 3 Paul Sabatier, F-31062 Toulouse, France

Search for more papers by this author
Jean-Marc Bonneville

Jean-Marc Bonneville

Lab Ecol Alpine LECA, Université Grenoble Alpes, F-38000 Grenoble, France

CNRS, Lab Ecol Alpine LECA, F-38000 Grenoble, France

Search for more papers by this author
Philippe Choler

Philippe Choler

Lab Ecol Alpine LECA, Université Grenoble Alpes, F-38000 Grenoble, France

CNRS, Lab Ecol Alpine LECA, F-38000 Grenoble, France

Search for more papers by this author
First published: 07 October 2015
Citations: 23

Summary

  • The effect of plant species composition on soil microbial communities was studied at the multiregional level. We compared the soil microbial communities of alpine natural grasslands dominated by Carex curvula and anthropogenic subalpine pastures dominated by Nardus stricta.
  • We conducted paired sampling across the Carpathians and the Alps and used Illumina sequencing to reveal the molecular diversity of soil microbes.
  • We found that bacterial and fungal communities exhibited contrasting regional distributions and that the distribution in each grassland is well discriminated. Beta diversity of microbial communities was much higher in C. curvula grasslands due to a marked regional effect. The composition of grassland-type core microbiomes suggest that C. curvula, and N. stricta to a lesser extent, tend to select a cohort of microbes related to antibiosis/exclusion, pathogenesis and endophytism.
  • We discuss these findings in light of the postglacial history of the studied grasslands, the habitat connectivity and the disturbance regimes. Human-induced disturbance in the subalpine belt of European mountains has led to homogeneous soil microbial communities at large biogeographical scales. Our results confirm the overarching role of the dominant grassland plant species in the distribution of microbial communities and highlight the relevance of biogeographical history.

Introduction

Due to the importance of soil microbiota on biogeochemical cycles and ecosystem processes and services (van der Heijden et al., 2008; Oates et al., 2012; Saleem & Moe, 2014), understanding the distribution patterns of microorganisms – that is, the changes in soil microbial diversity along spatial and environmental gradients – has been a focus of active research over the last years. For macro-organisms, it is now well accepted that the composition of local communities depends on multiple interconnected processes acting at different spatial scales (Thuiller et al., 2013). At large spatial scales, dispersal constitutes the first filter and relies on the organism's intrinsic dispersal capacity, its surrounding abiotic and biotic landscape (e.g. barriers or facilitators), and the biogeographical history. The resulting species pool is then subjected locally to niche filtering, which comprises the abiotic environment, species’ physiology and biotic interactions. Although there is now clear evidence that microbial biogeographical patterns are caused by dispersal limitation (Hanson et al., 2012), niche filtering effects (soil chemical properties, plant cover and land use) appear to be predominant (Fierer & Jackson, 2006; Griffiths et al., 2011; Martiny et al., 2011; Ranjard et al., 2013; Prober et al., 2015).

As primary producers, plants are the most important biotic filters for soil microbial communities. Plant community composition including plant species identity and local species abundance has also been shown to impact soil microbial communities at the local (Zinger et al., 2011; Rosenzweig et al., 2013), regional (Hovatter et al., 2011; Roy et al., 2013) and global scales (Tedersoo et al., 2014; Prober et al., 2015). The quality of soil organic matter, which is influenced by the dominant plants of the ecosystem, has been proposed to mediate this effect for saprotrophic microorganisms (see Millard & Singh, 2010). However, recent work also raises the importance of biotrophic interactions (Peay et al., 2013). Identification of soil microorganisms associated with the same plants across a range of spatial scales is hence central to understanding the interactions involved in microbial recruitment by the plants. Plant and soil feedbacks are complex (van der Putten et al., 2013) and produce confounding effects on the turnover of microbial communities. It is therefore difficult to disentangle the respective effects of plants and soils. Recent studies have focused on the plant-specific species on soil microbial communities (Talbot et al., 2014) or particular microbial functional groups (Martiny et al., 2011; Timling et al., 2012; Bahram et al., 2013). However, they have not addressed the relationships between microbial taxa and dominant plant species across large spatial scales – that is, their core community.

The core species hypothesis (Hanski, 1982) states that for a given habitat, core species are abundant on a local scale, and their populations are common at a regional scale. In microbial ecology, this concept was transposed from species to molecular operational taxonomic units (MOTUs). The core MOTUs, that is, those common to the same habitat (Turnbaugh et al., 2007), are thought to provide critical functions for the biotic assemblage (Shade & Handelsman, 2012). Their identification may shed light on how plants may recruit particular microbial species. Previous reports have so far addressed these questions for root endophytic bacterial communities but not for soil communities. A glasshouse work on Arabidopsis thaliana roots has shown the importance of soil microbial communities on the composition of root bacterial endophytic communities (Bulgarelli et al., 2012; Lundberg et al., 2012). A recent field study has also provided evidence that plants belonging to the same species harbour common microbial species in the rhizosphere (Rosenzweig et al., 2013). However, experiments on the roots of the perennial arctic Bistorta vivipara failed to identify a bacterial core community (Vik et al., 2013). To our knowledge, there has been no study examining core microbial species and plant filtering effects at large biogeographical scales.

The main goal of this work was to determine microbial biogeographical patterns across the Alps and the Carpathians as well as to identify the underlying processes amongst plant species assemblages and dispersal limitation. We used a paired sampling scheme to compare two plant species assemblages occurring on acidic soils in the temperate mountains of Europe. The first assemblage is the pristine alpine meadows dominated by the crooked sedge Carex curvula subsp. curvula – hereafter Ccu. The second is the anthropogenic subalpine grasslands dominated by the mat grass Nardus stricta – hereafter Nst. Ccu grasslands have a highly fragmented distribution and a long-term separation between populations from the Carpathian and the Alps, as inferred from the spatial genetic structure of C. curvula (Pușcaș et al., 2008a,b). By contrast, the subalpine Nst pastures are widespread, interconnected and more intensively used for pasturing (Hejcman et al., 2013). We also assume, at a scale encompassing several biogeographic regions, that the microbial species pools are different due to dispersal barriers and ecological drift (Martiny et al., 2011; Ranjard et al., 2013; Talbot et al., 2014). We therefore hypothesize that plant species primarily drive soil microbial community composition, that is that Ccu microbial communities are distinct from Nst ones. We nevertheless expect to find a certain degree of microbial regional endemism in Ccu, because these ecosystems have remained isolated and undisturbed for long periods as compared with Nst pastures. For the latter, we expect soil microbial communities to exhibit a lower differentiation across large spatial scales.

In order to test this hypothesis, soils of C. curvula and Nstricta grasslands were collected along three regions from an East–West transect across the Carpathians and the Alps (Fig. 1). We used DNA metabarcoding to study bacterial and fungal beta diversity – that is, community variation among sampling sites. More specifically, we investigated the respective influences of soil and plant assemblage composition on the distribution of soil microbial communities, determined which MOTUs were particular to each mountain region and/or grassland type, and delineated the respective core microbiomes of Ccu and Nst, as well as their presumed functions.

Details are in the caption following the image
Location of sampling sites. The colours indicate the Carex curvula (Ccu) (green) and C. curvula subsp. rosae (Cro) (red) grasslands and Nardus stricta (Nst) pastures (blue). Three mountain regions are defined: circles, Carpathians (CA); squares, Central and Eastern Alps (A1); diamonds, Western Alps (A2). Sites: FA, Făgăraș Mts; BU, Bucegi Mts; PA, Parâng Mts; RE, Retezat Mts; LA, Lavanttaler Alpen; GL, Glocknergruppe Alpen; OT, Ötztaler Alpen; BE, Alpes Bernoises; BA, Beaufortain-Aravis; TF, Massif du Taillefer; MO, Massif de Mercantour; CE, Massif des Cerces.

Materials and Methods

Study sites and characteristics

This study focuses on two types of subalpine/alpine grasslands that developed on acidic soils in the temperate mountains ranges of Europe, the so-called European Alpine System (Ozenda, 1985). The alpine grasslands dominated by Carex curvula (Ccu) have long been recognized as late-successional communities of the alpine belt on siliceous bedrocks (Niederfriniger-Schlag & Erschbamer, 2000; Pușcaș & Choler, 2012). In the Alps and the Carpathians, C. curvula grasslands exist above the potential treeline between c. 2200 and 2700 m above sea level. The pastures dominated by the mat-grass Nardus stricta (Nst) are anthropogenic grasslands that mostly occur in the subalpine belt (Kliment, 2007; Gennai et al., 2014). The extension of the N. stricta pastures in the European mountains has been promoted by forest logging and burning, followed by intensive grazing. The persistence of these grasslands is dependent upon land management. For example, the recent decreasing of grazing pressure in Europe has led to a sharp reduction of the surface occupied by mat-grass swards (Peppler, 1992; Korzeniak, 2009; Gennai et al., 2014).

In both types of plant communities and in most of the sampling sites, > 50% of the phytomass consists of the respective dominant species (Klug-Pümpel, 1982; Grabherr et al., 2000; Holland et al., 2008) (Supporting Information Fig. S1). It is therefore anticipated that the dominant grass and sedge have major implications on the structure and functioning of the ecosystems (Grime, 1998; Whitham et al., 2006).

In order to further assess the magnitude of differences between, and the possible molecular operational taxonomic units (MOTUs) shared by, the two different acidic grasslands, we used an ‘outgroup’ from a contrasted alpine habitat: the calcareous alpine grasslands dominated by C. curvula subsp. rosae (hereafter Cro). This taxon does not occur in the Carpathians. Bedrock and plant species assemblages of Cro grasslands significantly differ from those occurring in Ccu grasslands, and therefore represent a completely different type of habitat (Choler & Michalet, 2002; Choler et al., 2004). This comparison is needed to assess the MOTUs shared by Nst and Ccu, the acidic grasslands. Results including this outgroup are presented in the supplementary material only (Note S1).

Floristic surveys were conducted using standard phytosociological protocols (Braun-Blanquet, 1932). Briefly, the data consist in the relative cover of all vascular plant species present in homogeneous plots, usually from 25 to 100 m2. The relative cover was visually estimated on the basis of a 6-level scale: +, < 5%; 1, 5–10%; 2, 10–25%; 3, 25–50%; 4, 50–75%; and 5, > 75%. The resulting floristic table included a total of 196 species and is available at Dryad (http://datadryad.org/) under the identifier doi: 10.5061/dryad.ff138. We estimated dissimilarities in the floristic composition and structure between grasslands using the Bray–Curtis index.

Sampling

Soil sampling was performed between 17 July and 7 August 2007 in the Alps (hereafter regions A1 and A2 for East-Central Alps and Western Alps, respectively) and the Carpathians (region CA; Fig. 1; Table S1). At each sampling site, five (Nst) or 10 (Ccu) soil cores (15-cm deep) were randomly collected and sieved at 2 mm, kept at 4°C for > 8 d, and stored at −20°C until DNA extraction. Soil physicochemical properties were determined by the Laboratoire d'Analyse des sols (INRA, Arras, France) on dried soils (70°C for 72 h) using the following methods. Soil organic matter content (SOM) was estimated by loss-on-ignition with dry combustion (5 h at 550°C) with 15 g of starting material. Total carbon and nitrogen were measured by a Flash EA 1112 elemental analyser (Thermo Fisher Scientific Inc., Waltham, MA, USA) on 3–5 mg of sieved soil. Soil water pH was measured after mixing 5 g of soil with 12.5 ml of distilled water. The values are available at Dryad (http://datadryad.org/) under the identifier doi: 10.5061/dryad.ff138.

DNA isolation, PCR and sequencing

DNA was extracted from three aliquots of 0.25 g wet soil using the PowerSoil Soil DNA Isolation Kit (MO BIO Laboratories, Ozyme, St Quentin en Yvelines, France). The concentration of DNA extracts was quantified using the NanoDrop ND-1000 (NanoDrop Technologies, Wilmington, DE, USA). Bacterial communities were assessed with a primer pair targeting the V3–V4 regions of the 16S rRNA gene, rendering amplicons of c. 400 nt length. Primers were designed using the ecoPrimer program (Riaz et al., 2011) and RDPII database, resulting in the forward primer 5′ GCCAGCAGCCGCGGTAA 3′, denoted as F517 in (Liu et al., 2007), and the reverse primer 5′ GGACTACCAGGGTATCTAATC 3′, a shorter version of the primer 797R (Nadkarni et al., 2002). Fungal internal transcribed spacer 1 (ITS1) was amplified using the primers ITS5 5′ GGAAGTAAAAGTCGTAACAAGG 3′ (White et al., 1990) and 5′ 5.8S_fungi CAAGAGATCCGTTGTTGAAAGTT 3′ (Epp et al., 2012), rendering amplicons of c. 214 nt. All primers were extended by sample specific tags of 8 nt length to allow parallel sequencing of multiple samples. These tags were different for the forward and reverse primers. The PCR reactions were carried out in 25-μl reaction mixtures containing 10 ng of soil DNA, 0.25 mM of each primer, 0.2 mM of each dNTP, 2 mM MgCl2, 1 U of Ampli Taq Gold DNA polymerase and 2.5 μl of PCR buffer (Applied Biosystems, Courtabeuf, France). PCR reactions were run for 30 cycles at 95°C for 30 s, 56°C for 30 s and 72°C for 30 s, followed by a final extension step at 72°C for 10 min. The PCR products were purified with the QIAquick kit and quantified for DNA concentration using the QIAxcel system according to manufacturer's instructions (Qiagen). The purified amplicons were pooled for the subsequent sequencing by using equivalent molarities amongst amplicons. The library construction and sequencing (Illumina MiSeq 125-bp pair-end) were carried out at Fasteris (Geneve, Switzerland).

Bioinformatics filtering

The bioinformatics processing was performed using the ‘OBItools’ package (Boyer et al., 2015). The summary of sequence filtering and MOTU clustering is presented in Table S2. We aligned all paired-end reads with an algorithm that does not penalize end gaps. If the score of the resulting alignment was significantly higher than alignments generated with random reads, then the paired-end reads were assembled; otherwise, the pair-ended reads were joined bluntly. Reads were then dereplicated, and the singletons or unique sequences displaying ambiguous bases were removed. Unique sequences were then clustered into MOTUs, as previously reported (Zinger et al., 2009). Briefly, sequences were aligned pairwise with the ‘sumatra’ package (http://metabarcoding.org/sumatra). The resulting similarity matrix was then used to group sequences into MOTUs using the MCL (Markov Clustering) algorithm (van Dongen, 2000), with similarity thresholds of 98% for Fungi (Lentendu et al., 2011) and 97% for Bacteria (see Fig. S2 accumulation curves per sample or per grassland). Samples were rarefied to 660 reads per sample for Bacteria and 970 reads per sample for Fungi. These sample sizes correspond to samples with the lowest sequencing coverage minus 50. Hence, we produced a matrix of MOTU abundance per sample. The fasta files with the read counts per sample and per site are available at Dryad (http://datadryad.org/) under the identifier doi: 10.5061/dryad.ff138, and the most represented sequences of the first 1000 clusters are available at EMBL (EMBL accession no.: LN881743–LN883742). Fungal taxonomic assignment was performed as follows. First, we constructed an ITS1 reference database using the ecoPCR program (Ficetola et al., 2010). This consists of running an in silico PCR with the primer pair used here on the UNITE + INSDC database (release 24 September 2102) (Koljalg et al., 2013) so as to retain only the PCR-targeted region of sequences. Only reference sequences of 50–600 nt in length and with unambiguous taxonomic annotation were kept to enhance taxonomic assignments robustness. Second, global alignments for each sequence were then performed against this database using the ecotag script (Boyer et al., 2015). This script evaluates the last common ancestor of all references that are similar to the query sequence. This procedure ensures avoiding spurious taxonomic assignments such as those obtained on the basis of best matches only, which are strongly biased by taxonomic reference incompleteness and errors.

For bacterial taxonomic assignment, we used the naïve Bayesian Classifier (Wang et al., 2007) implemented in the RDPII tool ‘Classifier’ v2.6. Taxonomical assignation from next-generation sequencing has at least two major limitations critical for Fungi: the incompleteness of databases and limited sequence length (Hibbett & Taylor, 2013; Herr et al., 2015). For Bacteria, assignation is also not of high resolution due to the short sequence and the lesser variability of the 16S RNA gene. Note here that both taxonomic assignment tools provide taxa names in which we have enough confidence on the basis of the whole database composition. Finally, we associated a hypothetical lifestyle to each MOTU that had a fine-level taxonomic assignation according to the literature.

Statistical analysis

The rarefied read abundance data were used to produce an abundance matrix of MOTUs per sample used to identify significant associations of MOTUs to regions and/or grasslands. For multivariate analysis, we produced a MOTUs reads-by-site table to fit with floristic data. Soil chemical parameters were ordinated using a principal component analysis (PCA) and biotic assemblages were ordinated using a principal coordinated analysis (PCoA) (Legendre & Legendre, 2012) from Bray–Curtis distance matrices (Bray & Curtis, 1957). The correlations between biotic, geographic and soil chemistry distance matrices were tested using Mantel tests (Mantel, 1967), as defined by Legendre & Fortin (1989). We tested for significant associations between MOTUs and geography by identifying groups of sites that best explain the distribution patterns of microbial communities, following a method proposed by (Pușcaș et al., 2008a). The separation into three regions – Carpathians (CA), Central-Eastern Alps (A1) and Western Alps (A2) – provided the best results. Significant associations between MOTUs and subsets of regions were identified using χ2 tests (Pușcaș et al., 2008a) on the complete dataset (i.e. including Cro sites). This allowed us to identify MOTUs of the triple intersect. We then counted in each geographic region and/or grassland type the number of MOTUs above a double threshold of abundance (0.01%) and presence (at least five of the samples). Their specificity was validated by comparing their distribution against the null distribution predicted in a χ2 test using 1000 Monte Carlo resamplings. A false discovery rate (FDR) correction was applied to the raw P-values (Benjamini & Hochberg, 1995). MOTUs yielding a raw P-value lower than a corrected FDR P-value < 0.05 were retained. Finally, when a given MOTU was retained using several criteria, it was attributed in priority to the triple intersect (if present); otherwise, it was attributed to the area whose χ2 test gave the highest Pearson residue. For regions, MOTUs of the triple intersect are called here ‘acidic grassland MOTUs’ and represent the MOTUs present in all acidic grasslands but not in Cro. Those of the double intersects (i.e. present in two regions but not in Cro) or the relative complements (i.e. present only in the considered region) are called ‘endemic MOTUs’. For each grassland type, MOTUs of the triple intersect are those common to all regions for the considered grassland and absent from the other grassland and Cro. They are called hereafter the ‘core MOTUs’, while those of double intersects and relative complement are called ‘diagnostic MOTUs’. Statistical analyses were carried out using R software (R Development Core Team, 2013) and R packages ‘vegan’ (Oksanen et al., 2013), BiodiversityR (Kindt & Coe, 2005) and mvabund (Wang et al., 2012). The pool of species was estimated using the function specpool. Multivariate analysis of the deviance was performed with generalized linear models (manyglm) and tested with the function anova.manyglm.

Results

The chemical analysis of soils from Ccu grasslands and Nst pastures (Fig. S3) did not show any significant differences for pH or for nutrient contents (N, P and OM). Only the C : N ratio was higher for Ccu grasslands (Mann–Whitney U-test, P < 0.05). In the PCA ordination of soils variables, (Fig. 2a), the first axis was related to nutrients and the C : N ratio, while the second axis of variation was associated with pH. However, there was no clear discrimination of soil variables according to grassland type. Plant, bacterial and fungal communities were clearly separated along a gradient of C : N and pH, but the effect of soil chemistry was less important than those of geographic region or grassland type (Ccu or Nst; Fig. 2b–d). The plant species assemblages exhibited an E–W distribution in the ordination with a first clear grouping by mountain region and a second by grassland type (Fig. 2b), except for Ccu BA that corresponds to low-elevation alpine grasslands with a higher abundance of N. stricta (Fig. S1b) and other subalpine plant species. By contrast, microbial communities were first separated by grassland type; the clustering by geographic region was clearer for Ccu than for Nst, mainly because A2 sites cluster apart form CA and A1 (Fig. 2c,d). The longitudinal gradient was also observed for fungal communities.

Details are in the caption following the image
Ordination of (a) soil chemical properties, (b) plant, (c) bacterial and (d) fungal communities. (a) Principle component analysis (PCA) ordination of the soil chemical variables. Ccu grasslands are in green; Nst pastures are in light blue. P, phosphate; C, carbon : nitrogen ratio; OM, organic matter; OC, organic carbon; Nt, total nitrogen. (b–d) Principle coordinated analysis (PCoA) ordination of plant, bacterial and fungal communities, respectively. Soil chemical variables were on PCoA ordinations with envfit, and significant correlations (P-value < 0.05, FDR corrected) are indicated with black lines. Sites: FA, Făgăraș Mts; BU, Bucegi Mts;PA, Parâng Mts; RE, Retezat Mts; LA, Lavanttaler Alpen; GL, Glocknergruppe Alpen; OT, Ötztaler Alpen; BE, Alpes Bernoises; BA, Beaufortain-Aravis; TF, Massif du Taillefer; MO, Massif de Mercantour; CE, Massif des Cerces.

The correlation between biotic community dissimilarities, geographic distances and soil chemistry dissimilarities were further studied for Ccu grasslands and Nst pastures with Mantel tests (Fig. 3; Table S3). The biotic communities did not correlate with the soil chemical environment (Table S3). Fungal communities’ correlation with geographic distances was not significant when controlling for plant community dissimilarity (Table S3). Bacterial and fungal community dissimilarities correlated with plant community dissimilarities, and this correlation was significant even when controlling for geographic distances or soil chemistry differences among sites (Fig. 3; Table S3). We conclude that grassland type, and not geographical location or soil chemical environment, is the best predictor of bacterial and fungal community composition.

Details are in the caption following the image
Correlation between microbial and plant community dissimilarities. The Bray–Curtis dissimilarity of (a) bacterial and (b) fungal communities was plotted against the plant species assemblage for each site. Red lines indicate the linear fit between microbial and plant community dissimilarities.

We tested the respective influences of grassland type and previously identified geographic regions of Ccu grasslands (Pușcaș et al., 2008a; Pușcaș & Choler, 2012) on the composition of microbial communities by multivariate analysis of deviance (Table 1). Regions explained more variation than grassland type, but when assessing this effect for each grassland type separately, the variation of microbial communities explained by the regional factor was only significant for Ccu grasslands.

Table 1. Multivariate analysis of the deviance of generalized linear models fitting microbial communities from acidic grasslands to grassland type and region
Factor Dataseta Bacteria Fungi
Devb P-valueb Devb P-valueb
Grasslandc Ccu and Nst 1054 0.002 1870 0.001
Regiond 1814 0.001 2132 0.001
Region Ccu 908 0.018 1164 0.02
Region Nst 797 nse 907 nse
  • a The dataset analysed with mvabund.
  • b Contribution to the explained deviance and its significance after 1000 permutations in generalized linear models fits assuming a negative binomial distribution. The reported values correspond to a type II ANOVA table. P < 0.05 values are in bold.
  • c Grassland type: Ccu, Carex curvula; Nst, Nardus stricta.
  • d Mountain regions are Carpathians (CA), Central and Eastern Alps (A1) and Western Alps (A2).
  • e Not significant, P > 0.05.

We then used χ2 statistics to identify microbial MOTUs displaying preferential distribution in different regions and/or grassland type. For bacterial communities, we found 12 acidic grassland bacterial MOTUs out of 3279 MOTUs (Fig. 4), accounting for 12.5% of reads. Indeed, bacterial MOTUs that were abundant at one site were very often also abundant at most sites regardless of the type of grassland (data not shown). The highest numbers of bacterial endemic MOTUs were found for CA∩A1 and A2 regions. Ccu grasslands displayed six core MOTUs and Nst nine core MOTUs out of 2459 and 1752 MOTUs, respectively (Fig. 4; Table S4). In agreement with the presence of dominant acidic grassland MOTUs, core bacterial MOTUs were of lower abundance (Fig. 4). In most regions, Ccu grasslands displayed more diagnostic MOTUs than Nst pastures but also more than Ccu core MOTUs, except in A1∩A2 (Fig. 4). Interestingly, nine acidic grassland MOTUs were assigned to Acidobacteria, while the taxonomic affiliation of core bacterial MOTUs corresponded to endophytic and antibiotic-producer bacteria (data not shown). Taken together, these results indicate that bacterial communities of acidic grasslands were dominated by abundant species, while endemic, core and diagnostic MOTUs corresponded to low abundance taxa.

Details are in the caption following the image
Distribution of bacterial molecular operational taxonomic units (MOTUs) associated to region and to grassland type. MOTUs associated with a geographic region (Ccu + Nst, orange tones) and grassland test (Ccu or Nst, green and blue tones, respectively) were identified using the cognate χ2 test (see the Materials and Methods section). The upper panels show the percentage of MOTUs specific of each condition expressed as a percentage of the observed cognate richness. The lower panels show the cumulated abundance of MOTUs in the respective dataset. The observed richness of MOTUs was calculated using the command specpool. Ccu, Carex curvula; Nst, Nardus stricta; CA, Carpathians; A1, Central and Eastern Alps; A2, Western Alps.

The distribution of fungi across regions and grassland types was strikingly different from bacteria in four ways. First, we did not find any acidic grassland MOTUs (Fig. 5). Second, we identified more endemic MOTUs than for bacteria. These were more abundant in CA, A2 and CA∩A1 (Fig. 5). Third, Ccu harboured around six times fewer core MOTUs than Nst (seven and 41 MOTUs out of 1657 and 931 MOTUs, respectively) but displayed more diagnostic MOTUs, which were of higher abundance (Fig. 5). Finally, five out of the seven fungal core MOTUs in Ccu were clearly associated with biotrophic lifestyles (Table S5). Among the core MOTUs of Nst pastures exhibiting abundance higher than 0.1%, three were related to biotrophic clades (Trichoderma asperellum, Phialocephala sp. and Pseudeurotium bakeri; Table S5). Taken together, these results suggest a strong regional influence on fungal communities, which is reflected by a strong endemic background, especially in Ccu grasslands.

Details are in the caption following the image
Distribution of fungal molecular operational taxonomic units (MOTUs) associated to region and to grassland type. MOTUs associated with a geographic region (Ccu + Nst, orange tones) and grassland test (Ccu or Nst, green and blue tones, respectively) were identified using the cognate χ2 test (see the Materials and Methods section). The upper panels show the percentage of MOTUs specific of each condition expressed as a percentage of the observed cognate richness. The lower panels show the cumulated abundance of MOTUs in the respective dataset. The observed richness of MOTUs was calculated using the command specpool. Ccu, Carex curvula; Nst, Nardus stricta; CA, Carpathians; A1, Central and Eastern Alps; A2, Western Alps.

Discussion

We have studied the microbial communities of soils from two mountain grassland types that are distributed across the Alps and the Carpathians. There are four main results from this study. First, the dissimilarity between nearby sites differing in plant species assemblage was always much stronger than the dissimilarity within the same plant community, even if sampling sites were separated by hundreds of kilometres. Second, soil microbial communities in these grasslands were clearly different amongst regions. Third, the regional pattern was stronger in Carex curvula (Ccu) grasslands than in by Nardus stricta (Nst) pastures. Finally, fungal microbial communities of Nst pastures exhibited a richer core community. These observations confirm our hypothesis and open avenues in the phylogeography of microbial communities, as discussed later.

Influence of soil nutrients and grassland type

Soil chemical properties were poor predictors of microbial community dissimilarity, as expected from our pair-sampling strategy. Indeed, the C : N ratio was the only factor that differed significantly between these grassland types (Ccu being higher; Fig. S3). Lower C : N ratio is often associated with grazing, because herbivores return to the soil labile forms of C and N (Bardgett et al., 2001; Oates et al., 2012). Thus, the difference of C : N can be also linked to plant composition, highlighting the complex interactions between vegetation and soil. Plant species assemblages and grassland type were better predictors than geographic location or soil chemical properties (Figs 2, 3; Table S3). Plant community composition and grassland type are strongly dependent on the dominant plants. Indeed, Ccu and Nst are the local keystone species – that is, they are crucial for maintaining the diversity and organization of their ecosystem (Paine, 1969; Mills et al., 1993). Consequently, plant composition also correlates with other unmeasured parameters (e.g. insect or collembolan communities and other soil nutrients). From this point of view, our study confirms the important influence of plant community on soil microbial communities previously studied at larger spatial scales (Tedersoo et al., 2014; Prober et al., 2015).

Contrasted regional trends of microbial communities

We found clear regional patterns of microbial communities regardless of the type of grassland (Table 1). The trends were different for bacterial and fungal communities (Figs 2, 4, 5). Bacteria displayed little regional specificity, and acidic grasslands shared 12 abundant molecular operational taxonomic units (MOTUs) (Fig. 4). The presence of Acidobacteria points to the strong filtering exerted by acidic soil pH on bacterial communities of Ccu and Nst. Acidobacteria are more abundant in soils of low pH (Lauber et al., 2008) and display phylogenetic clustering (Bryant et al., 2008; Shahnavaz et al., 2012), which is indicative of selective pressure exerted by the abiotic conditins of the habitat. Nonetheless, we found few endemic bacterial MOTUs, albeit of low abundance, that explain the effect of region. We suggest that bacterial communities are more sensitive to the local abiotic environment (especially soil pH) than to biogeographic (regional) filtering. By contrast, fungal communities displayed a marked endemism (Fig. 5), which can be due to particular climatic preferences, limited dispersal, different regional pools of species or differences in biogeographic history. This is consistent with previous results at the landscape or global levels (Zinger et al., 2011; Prober et al., 2015). Also, regional endemism of fungal MOTUs was also reported in North American pine forests (Talbot et al., 2014).

Influence of grassland type and possible role of biogeographic history on core microbial communities

The grassland type influenced both bacterial and fungal communities but at different levels. The regional effect was statistically supported for Ccu grasslands only (Table 1). For bacteria, the strong selective effect of pH seemed to mask the effect of grassland type. But Ccu grasslands exhibited a consistent regional bacterial imprinting, that is, diagnostic MOTUs, that was weaker in Nst pastures (Fig. 4), even though core and diagnostic MOTUs were of low abundance in both cases. For fungal communities, grassland type and region imprints were much more pronounced (Fig. 5). Indeed we found a noticeable amount of fungal core MOTUs that exhibited a significant cumulated abundance. Moreover, diagnostic MOTUs were also found in Ccu grasslands, although they were less numerous and of lower yet noticeable cumulated abundance than Ccu core MOTUs.

Each grassland type displayed a particular pattern of core and diagnostic MOTUs. Nst pastures were more homogeneous across the examined range than Ccu grasslands – that is, they exhibited more core MOTUs, which were more abundant for fungal communities. It is conceivable that the strong pressure of grazing (reflected in the decrease of C : N ratio) resulted in the recruitment of specific microbial species, and/or that higher connectivity of the Nst pastures promoted the dispersal of particularly microbial species, leading to the establishment of a richer core community in Nst. Few or no diagnostic MOTUs were found in Nst, suggesting that the most active recruitment of microbial species is related to grazing. Nst pastures are secondary grasslands won by man over forests, in which N. stricta is left alone by wild or domesticated herbivores (Ellenberg, 1996; Jewell et al., 2005). Land-use is known to be an important driver of soil biotic communities: pastures in the Amazonian forest were recently found to host fungal communities more similar and homogeneous than primary or secondary forest (Mueller et al., 2014), suggesting that grazing has similar effects on such contrasted habitats.

The distribution of fungal Ccu diagnostic MOTUs was similar to those of endemic MOTUs (Fig. 5). These may be related to the long-term persistence of plant individuals, which may be over 2000–5000 yr (Steinger et al., 1996; de Witte et al., 2012), and low human activity. The pioneering individuals of C. curvula would have recruited microbes from the regional pool available at the colonization time, which could have been fixed by biotrophic or saprophytic interactions and subsequent persistence and proliferation of the plant. There is phylogeographic evidence that postglacial recolonization of the Ccu alpine habitat has proceeded by local vertical upward migration in the Carpathians and through an east–west recolonization wave in the Alps (Pușcaș et al., 2008a). These postglacial recolonization routes fit with the regionalism of Ccu soil microbial communities, supporting a prominent role of biogeographic history on the fungal beta diversity of Ccu grasslands. Furthermore, the A1–A2 split found for microbes mirrors the distribution of plant species and plant genotypic diversity on siliceous grasslands previously described (Thiel-Egenter et al., 2011). Our results suggest that following ice retreat, C. curvula recolonized available niches with a cohort of core biotrophic microbial species (the tiny core mycobiome; Fig. 5) and that recruited microbial species from the regional species pool, some of which may correspond to the current diagnostic MOTUs. Finally, the high number of endemic and Ccu diagnostic MOTUs in CA∩A1 (Figs 4, 5) suggests a historical link between these two regions. This is consistent with results from a clustering analysis of Ccu plant species assemblages, which highlighted floristic similarities between the easternmost part of the Alps and the Carpathians (Pușcaș & Choler, 2012).

Microbial functions associated with alpine grasslands

Our results point to an active role for plants in the recruitment of specific fungal species. It is not surprising that core microbial species reflect biotrophic plant–microbe interactions and that these effects are stronger for C. curvula and its fungal community. It has been proposed that the interaction of long-living plants such as C. curvula with pathogens leads to a reduction of the virulence, resulting in a long-lasting plant–pathogen interaction (van Molken & Stuefer, 2008). The fungal biotrophic lifestyle proposed here for the bulk of the Ccu and Nst core fungal MOTUs (Table S5) supports the hypothesis of specific microbial recruitment through endophytic and/or pathogenic interactions in selecting soil microbial species. In agreement with this idea, recent phylogeographic studies have suggested that microbial symbionts or pathogens follow macroorganisms during migration (Thompson et al., 2005; Criscione et al., 2006; Busby et al., 2012). For bacteria, the lack of taxonomic resolution prevented defining clear lifestyles for the core MOTUs (Table S4), but the presence of Actinoallomurus in Ccu core MOTUs suggests the recruitment of endophytic and/or antibiotic-producing clades (Janso & Carter, 2010; Indananda et al., 2011; Pozzi et al., 2011).

The most striking result presented here is the low cumulated abundance of Nst fungal diagnostic MOTUs when compared with endemic or Ccu diagnostic fungal MOTUs. This result again underlines the role of the biogeographic history and the influence of human activity on microbial assemblages. Microbial patterns could hence also be considered as an historical record, opening the possibility of using them to follow colonization or invasion patterns as well as to evaluate the influence of human activity. This result calls for greater consideration of biogeographic history when assessing the spatial distribution of microbial communities. Our results also highlight the importance of the regional species pool and of biotic interactions on the assembly of microbial communities. It is commonly thought that core microbial communities will reveal organisms that are crucial for the function of the community (Shade & Handelsman, 2012); instead, our results point to the importance of parasitic or pathogenic organisms within the core microbiomes.

Acknowledgements

This paper is dedicated to the memory of Professor Serge Aubert (1966–2015). We are grateful to J. Roy for helpful discussions and assistance with the last common ancestor algorithm, David I. Warton for his help with mvabund, Arnaud Foulquier for discussions on spatial correlation, and to M. Öpik (Editor) and the anonymous reviewers for their useful suggestions during the reviewing process. We are also indebted to R. Douzet, D. Rioux and C. Miquel who assisted in the field and in the laboratory. Logistical support was provided by the Station Alpine Joseph Fourier (Grenoble) and the A. Borza Botanical Garden (Babeș-Bolyai University in Cluj-Napoca). Part of this research was conducted on the long-term research site Zone Atelier Alpes, a member of the ILTER-Europe network. P.C., R.A.G. and J-M.B. are part of Labex OSUG@2020 (ANR10 LABX56). This work was supported by Agence Nationale de la Recherche (ANR) – France (Project ODYSSEE, ANR-13-ISV7-0004) and Executive Agency for the Financing of High Education, Research, Development and Innovation (UEFISCDI) – Romania (Project ODYSSEE, PN-II-ID-JRP-RO-FR-2012, no. 15/01.01.2014).

    Author contributions

    P.C. and R.A.G planned and designed the research. P.C., M.P. and R.A.G. conducted fieldwork. J-M.B., L.Z. and R.A.G. performed experiments. R.A.G. and P.C. analysed data. R.A.G. wrote the manuscript that was improved by L.Z., J-M.B., P.C. and M.P.