Modularity Reveals the Tendency of Arbuscular Mycorrhizal Fungi To Interact Differently with Generalist and Specialist Plant Species in Gypsum Soils

Abstract

Patterns in plant–soil biota interactions could be influenced by the spatial distribution of species due to soil conditions or by the functional traits of species. Gypsum environments usually constitute a mosaic of heterogeneous soils where gypsum and nongypsum soils are imbricated at a local scale. A case study of the interactions of plants with arbuscular mycorrhizal fungi (AMF) in gypsum environments can be illustrative of patterns in biotic interactions. We hypothesized that (i) soil characteristics might affect the AMF community and (ii) there are differences between the AMF communities (modules) associated with plants exclusive to gypsum soils (gypsophytes) and those associated with plants that show facultative behavior on gypsum and/or marly-limestone soils (gypsovags). We used indicator species and network analyses to test for differences between the AMF communities harbored in gypsophyte and gypsovag plants. We recorded 46 operational taxonomic units (OTUs) belonging to nine genera of Glomeromycota. The indicator species analysis showed two OTUs preferentially associating with gypsum soils and three OTUs preferentially associating with marly-limestone soils. Modularity analysis revealed that soil type can be a major factor shaping AMF communities, and some AMF groups showed a tendency to interact differently with plants that had distinct ecological strategies (gypsophytes and gypsovags). Characterization of ecological networks can be a valuable tool for ascertaining the potential influence of above- and below-ground biotic interactions (plant-AMF) on plant community composition.

INTRODUCTION

The close relationship between some plant species and particular geological substrates has long been recognized. Certain soil types can impose stressful conditions on plants due to low water and nutrient availability or high levels of heavy metals (dolomites, serpentines, gypsum), inducing adaptive syndromes in the plants growing on those soils (1). Plant species specifically adapted to particular stressful substrates are called “specialist” plants. Other plant species, called “generalist” plants, usually inhabit nonstressful substrates but can also inhabit stressful substrates. Plant communities from gypsum soils are composed of plant species defined as gypsophytes, or specialists, which occur only in soils with gypsum, and gypsovags, or generalists, which can occur in both gypsum soils and other soil types (2). Physiological and edaphic factors are responsible for the occurrence of gypsophytes and gypsovags, but there is no generalized adaptive strategy (3). Among these factors, the ability to accumulate sulfate (1), mineral nutrients (N, P, Ca, M, K), amino acids, and proteins (4) and a better water balance have been proposed. However, the factors controlling the distribution and occurrence of gypsophytes and gypsovags are still not fully understood (5), and in particular, the influence of above- and below-ground biotic interactions on the coexistence of gypsophytes and gypsovags has been largely unexplored.

Arbuscular mycorrhizal fungi (AMF) are responsible for establishing mycorrhizal symbioses with most vascular plants in all environments, including semiarid ecosystems, where AMF help plants to cope with nutrient deficiency, drought, salinity, and other stresses (6). There are case studies on the general abiotic versus biotic effectors of AMF community assembly (7, 8). Differences in soil type have been reported to be key factors determining AMF species and community composition (9); this is particularly relevant in stressful environments, such as serpentine soils (10–12), thermal soils (13), heavy metal soils, and saline soils (14, 15). However, some authors have pointed out that AMF communities might be more determined by biotic factors, such as the host plant (16, 17), or might even be more specific to plant functional groups than to individual plant species (18).

The evidence of the importance of AMF diversity for ecosystem functioning (19) has resulted in growing interest in identifying the species that colonize plants in natural ecosystems. Specifically, plant-AMF symbioses and AMF diversity have been studied recently in gypsum ecosystems, where it has been found that the AMF community is host plant dependent (20, 21), such that different AMF communities colonize perennial and annual plant species (22), and gypsovags exhibit a higher AMF infection rate than gypsophytes (5). Despite these previous findings, it still remains unknown whether the community composition of the AMF colonizing the roots of gypsovags differs markedly from that for gypsophytes when both are growing in gypsum soils.

Recently, network analysis has been applied to characterize AMF communities and plant-AMF interactions (23, 24). Network analyses allow the detection of generalized patterns within the structure of biotic interactions at the community level. Network modularity reflects the tendency of a set of species to interact predominantly with species within the set and less frequently with species in other sets. Modularity in these networks implies that there are distinct communities of plants and AMF (modules) that interact more among themselves than with other plants or AMF.

In this study, we used indicator species analysis and network analyses to characterize and test for differences in AMF communities harbored in gypsophytes in gypsum soils and in gypsovags in gypsum and nongypsum soils. Subsequently, we explored the factors that best explain the grouping of species within a module (communities). We hypothesized that soil characteristics might affect the AMF community that can inhabit gypsum soils, resulting in different AMF communities (modules) in gypsum and nongypsum (marly-limestone) soils. Also, focusing only on gypsum soils, we hypothesized that there are differences between the AMF communities (modules) associated with gypsophyte and gypsovag plants. This study considers the potential influence of above- and below ground biotic interactions (plant-AMF) on plant community composition.

MATERIALS AND METHODS

Study area and plant sampling.

The study area was located in Sierra de Las Ventanas, Albatera, Alicante, in southeastern Spain (38°14′N, 0°55′W; altitude, 320 m). The climate is semiarid, with annual evapotranspiration (ETP) of 954 mm, annual average rainfall of 290 mm, a pronounced dry season from June to September, and a mean annual temperature of 18.4°C. It is a natural zone with gypsum outcrops and marly-limestone soils in a patchy distribution; the vegetation is scrubland (woody perennial species) dominated by gypsophytes in the gypsum areas and by typical scrub species of semiarid plant communities in the adjacent soils.

The soils are classified as Petrogypsid (25), with a gypsic and petrogypsic horizon within 100 cm of the surface, developed on gypsum parental rocks, and as Typic Torriorthent, with low organic matter content (Table 1).

TABLE 1.

Biological, chemical, and physical characteristics of the soil in the two areas sampled

Characteristica	Value (mean ± SE)b
	Gypsum area	Marly-limestone area
Alkaline phosphatase (μmol PNP g−1 h−1)	0.89 ± 0.12 A	0.31 ± 0.06 B
Dehydrogenase (μg INTF g soil−1)	30.66 ± 2.60 A	37.66 ± 4.85 A
Protease (μmol NH4+ g−1 h−1)	0.09 ± 0.02 A	0.14 ± 0.03 A
β-Glucosidase (μmol PNP g−1 h−1)	0.18 ± 0.02 A	0.13 ± 0.03 A
Available P (mg kg−1)	1.27 ± 0.4 B	4.96 ± 0.3 A
EC (μS cm−1)	2.400 ± 100 A	2.380 ± 100 A
pH	7.9 ± 0.3 A	8.3 ± 0.2 A
TOC (g kg−1)	5.5 ± 5.0 A	4.2 ± 5.0 A
Total N (g kg−1)	0.7 ± 0.2 A	0.6 ± 0.2 A
WSC (μg g−1)	68 ± 6 A	67 ± 4 A
GRSP (μg g soil−1)	31.66 ± 12.46 B	85.96 ± 8.15 A
CaCO3 (g kg−1)	216 ± 53 B	506 ± 72 A
CaSO4·2H2O (g kg−1)	434 ± 35 A	21.8 ± 8 B

^aINTF, iodonitrotetrazolium formazan; EC, electrical conductivity; TOC, total organic carbon; WSC, water-soluble C; GRSP, glomalin-related soil protein.

^bFor 5 samples per area. Values in the same column followed by the same letter do not differ significantly (P, >0.05) as determined by the Duncan test.

The plant species sampled were classified into two groups according to their ecological strategies. On the one hand, three of the most widely distributed gypsophytes (26) were considered: Ononis tridentata L., Helianthemum squamatum (L.) Dum. Cours., and Launaea pumila (Cav.) O. Kuntze. On the other hand, three species common throughout Mediterranean semiarid areas were considered as gypsovags: Globularia alypum L., Helichrysum stoechas (L.) Moench, and Anthyllis terniflora (Lag.) Pau.

Ten plots (2 by 5 m [10 m²]) were established randomly in the experimental area (with at least 5 m between plots), five with gypsum outcrops and five with marly-limestone outcrops. Within each plot, three individual plants for each selected plant species were sampled. Ninety plants (45 gypsophytes and 45 gypsovags) were sampled in the gypsum area and 45 plants (gypsovags) outside the gypsum area, for a total of 135 root samples. At the same time, one soil sample per plot (depth, 0 to 20 cm) was collected for soil characterization.

All samples were collected in the second half of May (late spring). Plants, including root systems, were collected, placed in polyethylene bags, and transported to the laboratory, where fine roots were separated from the soil. The roots were then briefly rinsed, quickly dried on paper, and used for molecular analysis.

Soil analysis.

Alkaline phosphatase activity was determined using p-nitrophenyl phosphate disodium (PNPP; 0.115 M) (Fluka) as the substrate. The p-nitrophenol (PNP) formed was determined by spectrophotometry at 398 nm (27).

β-Glucosidase activity was determined using p-nitrophenyl-β-d-glucopyranoside (PNG; 0.05 M) as the substrate. This assay is based on the release and detection of PNP. The amount of PNP was determined at 398 nm (28).

Dehydrogenase activity was determined according to the methods of Garcia et al. (29) and Trevors (30).

N-Benzoyl-l-argininamide (BAA)-hydrolyzing protease activity was determined in 0.1 M phosphate (Panreac) buffer at pH 7; 0.03 M BAA (MP Biomedicals) was used as the substrate. The activity was determined as the amount of NH₄⁺ released in the hydrolysis reaction (31).

Soil pH and electrical conductivity were measured in a 1:5 (wt/vol) aqueous solution. In aqueous extracts of soil, the concentration of water-soluble carbon was determined in an automatic carbon analyzer for liquid samples (TOC-V CSN analyzer; Shimadzu).

The glomalin-related soil protein (GRSP) concentration was determined in the easily extractable glomalin form, according to the method of Wright and Anderson (32).

Total N and total organic carbon concentrations were measured with a FlashEA 1112 elemental analyzer (Thermo).

Available phosphorus was extracted with 0.5 M NaHCO₃ (1:10, wt/vol) for 30 min and was measured colorimetrically.

The carbonate content was determined by Bernard calcimetry, and the percentage of gypsum in the soil was estimated by measuring total S with an elemental analyzer (LECO).

Extraction of DNA from roots and PCR.

For each root sample, 0.1 g fresh root material was frozen with liquid nitrogen, placed in a 2-ml screw-cap propylene tube together with two tungsten carbide balls (3 mm), and ground (3 min, 13,000 rpm) using a mixer mill (MM 400; Retsch, Haan, Germany). Total DNA was extracted by using a DNeasy Plant minikit according to the manufacturer's recommendations (Qiagen). The extracted DNA was resuspended in 20 μl of water.

DNA was extracted from 135 root samples. The three DNA samples extracted from each plant species within a plot were pooled into one composite sample, resulting in a total of 45 samples. This procedure was performed to ensure maximum coverage of the inter- and intraspecific variation within each replicated plot.

Several dilutions of the extracted DNA (1/10, 1/50, 1/100) were prepared, and 2 μl was used as a template. Partial small-subunit (SSU) rRNA gene fragments were amplified by using nested PCR with the universal eukaryotic primers NS1 and NS4 (33). PCR was carried out in a final volume of 25 μl, using Ready-To-Go PCR beads (Amersham Pharmacia Biotech), 0.2 μM deoxynucleoside triphosphates (dNTPs), and 0.5 μM each primer. PCR conditions were as follows: 94°C for 3 min; 30 cycles at 94°C for 30 s, 40°C for 1 min, and 72°C for 1 min; and a final extension period at 72°C for 10 min.

Two microliters of the 1/10-diluted product from the first PCR was used as the template DNA in a second PCR performed using the specific primers AML1 and AML2 (34). The PCRs were carried out in a final volume of 25 μl, using PuReTaq Ready-To-Go PCR beads (Amersham Pharmacia Biotech), 0.2 μM dNTPs, and 0.5 μM each primer. PCR conditions were as follows: 94°C for 3 min; 30 cycles consisting of 1 min of denaturation at 94°C, 1 min of primer annealing at 50°C, and 1 min of extension at 72°C; and a final extension period of 10 min at 72°C. Positive and negative controls (PCR positive products and sterile water, respectively) were also included in all amplifications. All the PCRs were run on a Perkin-Elmer/Cetus DNA thermal cycler. Reaction yields were estimated by using a 1.2% agarose gel containing GelRed (Biotium).

Cloning and sequencing.

The PCR products with the expected band, approximately 795 bp, were purified using a gel extraction kit (Qiagen), cloned into pGEM-T Easy (Promega), and transformed into Escherichia coli (XL1-Blue). From the 45 clone libraries, a total of 1,440 clones were screened by PCR. Thirty-two positive transformants were screened in each resulting SSU rRNA gene library, using 0.7 U of REDTaq DNA polymerase (Sigma) and reamplification with primers AML1 and AML2, under the same conditions as those described above. Product quality and size were checked in agarose gels as described above. Insert-containing clones (n, 1,019) were sequenced.

Clones were grown in liquid culture, and the plasmid was extracted using the QIAprep Spin Miniprep kit (Qiagen). Sequencing was carried out by the Laboratory of Sistemas Genómicos (Valencia, Spain) using the universal primers SP6 and T7. Sequence editing was done using the Sequencer program, version 4.1.4 (Gene Codes Corporation).

Phylogenetical analysis.

Sequence similarities were determined using the Basic Local Alignment Search Tool (BLASTn) sequence similarity search tool (35), provided by NCBI. Phylogenetic analysis was carried out on the sequences obtained in this study and those corresponding to the closest matches from the GenBank and MaarjAM (36) databases, as well as on sequences from cultured AMF taxa, including representatives of the major groups of Glomeromycota from GenBank. Sequences were aligned using the ClustalX program (37), and the alignment was adjusted manually with BioEdit software, version 7.0.4.1 (38). Neighbor-joining (NJ) (39) and maximum likelihood (ML) phylogenetic analyses were performed with the programs PAUP4.08b (40) and RAxML, version 7.0.4 (41), respectively. Distances for the NJ tree were computed using the default parameters. For the ML analysis, a GTR-Gamma model (general time-reversible model with gamma-distributed rates) of evolution was used. A total of 200 independent bootstrap analyses were performed to provide nodal support. The ML bootstrap values were calculated with 1,000 replicates, using the same substitution model. Endogone pisiformis Link and Mortierella polycephala Coem. were used as the outgroups.

Different sequence types, or operational taxonomic units (OTUs), were defined as groups of closely related sequences, usually with a high level of bootstrap support in the phylogenetic analyses (>80%) and with pairwise similarity (>97%). Pairwise analysis within clusters was carried out using BioEdit software, version 7.0.4.1 (38).

Statistical analysis.

For each plant species in each plot, the number of sequences and the number of samples for each AMF OTU was used to construct the sampling effort curves (with 95% confidence intervals) with EstimateS software, version 8.00. The sample order was randomized by 100 replications.

We used indicator species analyses to generate a numerical classification of OTUs (42). This method uses a reciprocal averaging ordination to classify the OTUs according to apparently important environmental properties (43). We performed two indicator species analyses, using the soil type and plant species, respectively, as the ordination properties.

The indicator value (IndVal) index (44) was used to measure the association between a species and a site group. Finally, the statistical significance of this relationship was tested using a permutation test with 999 permutations. We performed these analyses using the “indcspecies” package implemented in R (45).

Modularity.

Plant-AMF interactions were characterized by considering that there is a link between a plant and an AMF taxon (i.e., interaction) if an AMF OTU is present in the roots of a given plant species. This qualitative matrix (with a score of 0 or 1) was used to calculate network modularity.

The nodes (i.e., plant species and AMF OTUs) of a network can be grouped into modules, in such a way that the number of links (i.e., the presence of an interaction) within a module is maximized and the number of links between modules is minimized. We used a modularity algorithm for unipartite networks to search for independent groups of AMF that tend to be harbored in the same plants (i.e., that share a similar interaction pattern) (see Olesen et al. [46] for a more detailed explanation). In this type of network, an interaction between two AMF OTUs will be present if they share at least one host (i.e., plant). A simulated annealing optimization approach was used to detect modules that maximized modularity (i.e., the proportion of links within versus between modules) and to test if the pattern observed differed significantly from randomness, using 100 randomizations (47, 48). Because of the heuristic nature of the algorithm, 10 runs were conducted, but the variation in modularity was negligible (the standard error [SE] of the modularity across the 10 runs ranged from 0.0104 to 0.0106 when all interactions were considered and from 0.0113 to 0.0115 when only plants growing in gypsum soils were considered). We report the maximum value of modularity obtained in the 10 runs. The modularity was calculated and its significance tested using NetCarto software (47–49).

First, we tested for the contribution of the soil type (i.e., gypsum and nongypsum soil) to explain the assignment of each AMF OTU to a given module. Each OTU-plant interaction was classified based on the module to which the OTU was ascribed and the soil in which the plant was growing. We used multinomial regression models with the module as the dependent variable and the soil type as the independent variable. The model was compared with a null model in which the independent variable was a constant. As a rule of thumb, if the proposed model's Akaike information criterion (AIC) index is more than 2 units lower than that of the null model, it has some support, and if the AIC value is more than 10 units lower, it has substantial support (50). Second, we arranged the data set into subsets, selected the interactions that occur only in the gypsum area, and recalculated the modularity. Using the same methodology described above, we built multinomial regression models to explore the relative contribution of plant ecological strategy (gypsophytes or gypsovags) to the assignment of each OTU to a given module, considering only the plant-AMF associations occurring within the gypsum soil.

Nucleotide sequence accession numbers.

The 135 unique sequences of the clones generated in this study have been deposited at the National Center for Biotechnology Information (NCBI) GenBank (http://www.ncbi.nlm.nih.gov) under accession numbers HF559237 to HF559371.

RESULTS

Soil analysis.

The gypsum and marly-limestone soils had similar fertility levels, except for the available phosphorus content, which was significantly higher in the marly-limestone soil. Biological properties (enzymatic activities) did not differ significantly between the two soil types. The most marked differences were the significantly larger amounts of calcium sulfate in the gypsum soil and of calcium carbonate in the marly-limestone soil (Table 1).

PCR and sequence analysis.

The BLAST search revealed that 734 sequences (72%) had a high degree of similarity (94 to 100% similarity) to sequences from taxa belonging to the phylum Glomeromycota. The rest of the sequences were attributed to Ascomycota and plants.

Phylogenetic analysis of AMF OTUs.

The phylogenetic analyses of 734 glomalean sequences obtained in this study from plant roots, and of 50 sequences downloaded from GenBank, made possible the recognition of 46 OTUs as separate clades, on the basis of bootstrap values of ≥80%. According to Schüßler and Walker (51), the sequence groups or OTUs covered six families of the Glomeromycota: the Glomeraceae, Claroideoglomeraceae, Diversisporaceae, Paraglomeraceae, Gigasporaceae, and Archaeosporaceae. The pairwise sequence similarities within the clades ranged from 97.5 to 100%. The 135 clones that produced unique sequences were represented in the NJ and ML trees. Of all the AMF OTUs, 26 belonged to the genus Glomus, 2 to Funneliformis, 6 to Claroideoglomus, 1 to Sclerocystis, 1 to Rhizophagus, 2 to Diversispora, 1 to Redeckera, 1 to Scutellospora, 1 to Archaeospora, and 5 to Paraglomus (Fig. 1).

FIG 1.

Neighbor-joining phylogenetic tree showing AMF sequences isolated from the roots of Ononis tridentata, Helianthemum squamatum, Launaea pumila, Globularia alypum, Helichrysum stoechas, and Anthyllis terniflora, as well as reference sequences from GenBank. All bootstrap values of >80% are shown (1,000 replicates). Numbers above branches indicate the bootstrap values of the maximum likelihood analysis. Sequences obtained in the present study are shown in boldface. They are labeled with the host plant from which they were obtained (OT, O. tridentata; HS, H. squamatum; LP, L. pumila; GA, G. alypum; ST, H. stoechas; AT, A. terniflora) and the clone identity number. The letter G before a gypsovag species name means that it was obtained from the gypsum soil. Group identifiers (for example, G1) are used for OTUs found in our study. See the supplemental material for a detailed description of all the clones obtained in the present study for each group. Endogone pisiformis and Mortierella polycephala were used as outgroups.

AMF OTU distribution.

To determine whether the number of clones sequenced and the number of samples taken were sufficient to represent the diversity of the AMF in the roots, sampling effort curves were constructed (see Fig. S1 and S2 in the supplemental material). The results indicate that the number of sequences analyzed and the number of samples were sufficient to provide coverage of the AMF diversity in roots from the six plant species studied. In fact, all the curves reached the plateau.

The richness of the OTUs found in the roots of the different plant species in each soil type is shown in Table 2.

TABLE 2.

Number of sequences of each OTU and richness per plant species and per soil type

OTU	No. of sequences in:									Total no. (%) of sequences
	Gypsum soil						Marly-limestone soil
	Gypsophytes			Gypsovags
	Launaea pumila	Ononis tridentata	Anthyllis terniflora	Globularia alypum	Helianthemum squamatum	Helichrysum stoechas	Anthyllis terniflora	Globularia alypum	Helichrysum stoechas
P1			2							2 (0.27)
P2							2			2 (0.27)
P3							2			2 (0.27)
P4			2							2 (0.27)
P5		2	2	6			14	10	8	42 (5.72)
Ar1					2					2 (0.27)
ScuI				2						2 (0.27)
Re1						2				2 (0.27)
Cl1						2				2 (0.27)
Cl2		2	2				2		2	8 (1.09)
Cl3			2							2 (0.27)
Cl4								2		2 (0.27)
Cl5			2	2	2			2	9	17 (2.32)
Cl6		2					2			4 (0.54)
G1	14	6	5	13	7	2	10	3	26	86 (11.72)
G2				2			2	2		6 (0.82)
G3	2									2 (0.27)
G4				2						2 (0.27)
G5					2					2 (0.27)
G6				2	2			2	2	8 (1.09)
G7									3	3 (0.41)
G8		3					12	14	13	42 (5.72)
G9	8	7	4	2	2					23 (3.13)
G10			2	18	6	7				33 (4.5)
G11				2						2 (0.27)
G12	2			2			3		2	9 (1.23)
G13		2							2	4 (0.54)
G14	2		2			3	2			9 (1.23)
G15		5	2	2	2			7	3	21 (2.86)
G16									2	2 (0.27)
G17	4		2							6 (0.82)
G18	2	5				2		4	5	18 (2.45)
G19			2	2						4 (0.54)
G20		6								6 (0.82)
G21					2					2 (0.27)
G22		4	2	3			2		2	13 (1.77)
G23				2						2 (0.27)
G24	47	6	14	16	11	24	4	2	24	148 (20.16)
G25	16	13	34	5	15	26	6	2	13	130 (17.71)
G26					2					2 (0.27)
D1			2							2 (0.27)
D2				2				2	2	6 (0.82)
Fu1			2				2			4 (0.54)
Fu2	2	2	7		2		5		5	23 (3.13)
S1				2						2 (0.27)
Rh1		2	10				4	2	3	21 (2.86)
Total	99	67	102	87	57	68	74	54	126	734 (100)
Richnessa	10	15	20	19	13	8	16	13	18

^aTotal OTUs.

The most abundant OTUs in this study were Glomus G24 (20.1% of the clones belonged to this OTU), Glomus G25 (17.7%), and Glomus G1 (11.7%), which occurred in all the plant species, followed by Glomus G8 (5.7%) and Paraglomus P5 (5.7%), present in the roots of all the plant species growing outside the gypsum soil.

Indicator species analysis was conducted to find specific OTUs for the host and soil types. Two OTUs tended to occur in gypsum soils, G10 and G9 (IndVal, 0.66 [P, 0.007] and 0.58 [P, 0.013], respectively) and three in nongypsum (marly-limestone) soils, G8, Cl5, and P5 (IndVal, 0.64 [P, 0.003], 0.57 [P, 0.023], and 0.57 [P, 0.035], respectively) (Table 3). Regarding plant species, just the OTU G20 was significantly associated with O. tridentata (IndVal, 0.78 [P, 0.003]) (Table 4).

TABLE 3.

Indicator species analysis by soil type

Group (no. of species)	OTU	Indicator value index	P valuea
Gypsum soil group (24)	G10	0.658	0.007**
	G9	0.577	0.013*
	G24	0.689	0.107
	G25	0.648	0.189
	G17	0.316	0.520
	G20	0.316	0.536
	Re1	0.258	0.551
	G19	0.258	0.575
	G14	0.304	1.000
	G3	0.183	1.000
	G4	0.183	1.000
	G5	0.183	1.000
	G12	0.183	1.000
	G21	0.183	1.000
	G23	0.183	1.000
	G26	0.183	1.000
	Cl1	0.183	1.000
	Cl3	0.183	1.000
	S1	0.183	1.000
	D1	0.183	1.000
	Ar1	0.183	1.000
	D1	0.183	1.000
	P1	0.183	1.000
	P4	0.183	1.000
Marly limestone soil group (22)	G8	0.639	0.003**
	Cl5	0.566	0.023*
	P5	0.566	0.035*
	G7	0.365	0.108
	D2	0.414	0.109
	Fu1	0.516	0.169
	G2	0.327	0.263
	Fu2	0.327	0.264
	P2	0.258	0.301
	Cl4	0.258	0.340
	P3	0.258	0.343
	G1	0.617	0.348
	G16	0.258	0.371
	Rh1	0.405	0.442
	G6	0.298	0.564
	G11	0.298	0.580
	Cl2	0.298	0.611
	G15	0.390	0.716
	G18	0.330	1.000
	G22	0.276	1.000
	G13	0.211	1.000
	Cl6	0.211	1.000

^a**, P < 0.01; *, P < 0.05.

TABLE 4.

Indicator species analysis by host

Host group (no. of species)	OTU	Indicator value index	P valuea
Gypsovags
Anthyllis cytisoides group (13)	Fu2	0.548	0.112
	G14	0.447	0.125
	Rh1	0.500	0.142
	P5	0.400	0.245
	G22	0.283	0.447
	G14	0.316	0.865
	Cl3	0.316	1.000
	D1	0.316	1.000
	P1	0.316	1.000
	P2	0.316	1.000
	P3	0.316	1.000
	P4	0.316	1.000
	G19	0.224	1.000
Globularia alypum group (8)	G2	0.365	0.288
	D2	0.316	0.391
	G4	0.316	1.000
	G12	0.316	1.000
	G23	0.316	1.000
	Cl4	0.316	1.000
	S1	0.316	1.000
	D1	0.316	1.000
Helichrysum stoechas group (6)	Cl1	0.500	0.153
	G7	0.447	0.164
	Re1	0.447	0.173
	G1	0.391	0.853
	G16	0.316	1.000
	Cl1	0.316	1.000
Gypsophytes
Helianthemum squamatum group (6)	G10	0.474	0.118
	G5	0.447	0.326
	G21	0.447	0.326
	G26	0.447	0.326
	Ar1	0.447	0.326
	G6	0.283	0.758
Launaea pumila group (4)	G17	0.566	0.073
	G3	0.447	0.328
	G24	0.416	0.681
	G11	0.283	0.742
Ononis tridentata group (9)	G20	0.775	0.003**
	G15	0.507	0.106
	G9	0.474	0.233
	G8	0.381	0.367
	G18	0.365	0.569
	Cl6	0.365	0.605
	G13	0.365	0.615
	G30	0.283	0.773
	G25	0.405	0.929

^a**, P < 0.01.

Modularity.

The networks showed significant modularity, considering all interactions (modularity, 0.152 [confidence interval of the null model {CIN}, 0.105 to 0.107]) or considering only plants growing in the gypsum area (modularity, 0.185 [CIN, 0.092 to 0.137]) (Table 5). Four modules were detected in both networks, the one considering all the interactions and the other considering only interactions occurring in gypsum soil (Table 5). When all the interactions were considered, the ascription of OTUs to different modules was explained significantly by the soil type (AIC value for the null model, 632.6; AIC value using soil type as an explanatory factor, 613.8). The AMF OTUs ascribed to module B (Table 5) tended to prefer the gypsum area (multinomial regression coefficient, 0.13), and the OTUs ascribed to modules C and D tended to strongly avoid the gypsum area (i.e., they preferred the nongypsum area), compared to a reference module (module A [Table 5]) (multinomial regression coefficients, −1.43 and −1.06, respectively).

TABLE 5.

Modularity values of the unipartite AMF network^a

Network	OTUs	Modularity (CIN)
All interactions		0.152 (0.105–0.107)
Module A	G24, G12, G19, P4, S1, P2, G25, Re1, G2, Cl1, D1, G3, G4, De1
Module B	G10, G23, G9, G6, Ar1, Fu1, G5, G26, G21
Module C	G11, G14, Cl5, G17, P1, Cl3, G22, Fu2
Module D	G7, G16, G18, G13, D2, G8, G15, Cl4, Cl6, P3, G2, Cl2, P5, Rh1, G1, G20
Gypsum area		0.185 (0.092–0.137)
Module E	D2, G11, G9, G23, G1, G10, G21, G26, G5, Ar1, G6
Module F	Fu2, Cl3, Rh1, P1, G22, P5, G17, G14, Cl5
Module G	Cl2, Cl6, G12, G15, G20, G8, Fu1, G18, G13
Module H	G24, Re1, Cl1, S1, P4, G19, G4, G2, De1, D1, G3, G25

^aModularity values and confidence intervals are presented for each subset of data considering (i) all interactions and (ii) only plants growing in the gypsum area. The assignment of AMF OTUs to each module is presented for each analysis. Four modules (A to D and E to H) were detected in both data sets.

Within the gypsum area data set, plant ecological strategy contributed significantly to the explanation of the ascription of AMF OTUs to different modules (AIC value for the null model, 418.10; AIC value using plant ecological strategy as an explanatory factor, 408.90). The OTUs ascribed to module G (Table 5) tended to avoid gypsovags (i.e., they prefer gypsophytes) (multinomial regression coefficient, −1.22) more than the OTUs assigned to the reference module (module E [Table 5]). Meanwhile, the OTUs ascribed to modules F and H (Table 5) tended to associate more with gypsovags (multinomial regression coefficients, 0.98 and 0.31, respectively).

DISCUSSION

This study recorded 46 OTUs, belonging to nine genera (Archaeospora, Glomus, Diversispora, Claroideoglomus, Funneliformis, Sclerocystis, Rhizophagus, Redeckera, Scutellospora, and Paraglomus). This level of richness is higher than that found by a previous study in different gypsum areas (21). However, the results may not be directly comparable, since different primers were used. In the previous study, the NS31/AM1-3 primers developed by Santos-González et al. (52) were used; these cover a shorter SSU rRNA gene fragment than the AML1/AML2 primers of Lee et al. (34). In fact, no sequences from the present trial showed either identical or similar (97%) homology with any sequences obtained in the earlier study. This greater richness can also be attributed to the number of plant species sampled: six species, three of them in two different soil types.

The most abundant OTUs in this study were Glomus G24 (20.1% of the clones belonged to this OTU), Glomus G25 (17.7%), and Glomus G1 (11.7%), which occurred in all the plant species. They clustered with root-derived sequences of uncultured AMF species, and they have been recorded in other semiarid ecosystems (20, 21). OTUs G24 and G25 are closely related to the Rhizophagus intraradices group (sequence homology, 96%), which represents a ubiquitous generalist fungus, since it is one of the most common taxa and has been found in a broad range of environments (36, 53); Öpik et al. (53) proposed that some AMF species occur globally, showing high local abundance and low specificity, and the R. intraradices group clearly falls into this category as a generalist species. The presence of “potential specialist” AMF in gypsum ecosystems has been suggested (21). In the present work, the indicator species analysis showed two OTUs (G10 and G9) that tended to occur in gypsum soils and three OTUs (G8, Cl5, and P5) that tended to occur in marly-limestone soils. However, further studies are required before one can assert that certain AMF tend to occur in these types of soil. Furthermore, OTUs G10 and G9 have been reported previously by Wubet et al. (54, 55) in nongypsum soils from tropical montane forests in Ethiopia; still, these OTUs may represent peculiar ecotypes adapted to the particular environmental conditions of gypsum soils.

A major environmental factor producing differences in the AMF community composition in our work was related to the soil type. The modularity analysis supported this idea. There are indications that AMF diversity and distribution in soils depend on soil properties such as water availability (56), soil texture (57), or soil chemistry (14, 15, 58). Oehl et al. (9), in an extensive study, found that soil type is a key factor determining the composition of the AMF community. Recent AMF diversity studies have revealed that the AMF community composition is host plant dependent (16, 17, 21, 22, 59–64). Interestingly, when we focused on the plant and AMF communities inhabiting gypsum soils, we also found differences between the AMF harbored in gypsophyte and gypsovag plants. Similarly, Öpik et al. (18) found, in a boreal ecosystem, that AMF communities are more specific to plant functional groups than to individual plant species. The modularity and multinomial regression model results support the hypothesis that the soil type structures the AMF community composition. We found OTUs ascribed to modules preferring plants in the gypsum areas (module B [Table 5]) and others ascribed to modules preferring plants in the nongypsum areas (modules C and D [Table 5]). However, when we focused on the gypsum areas, particular patterns emerged. Within the gypsum areas, there were OTUs ascribed to modules with a tendency to interact with gypsophytes (module G [Table 5]), other OTUs ascribed to modules with a tendency to interact with gypsovags (modules F and H [Table 5]), and others that did not show a preference for interaction with either of the two types of plants (module E [Table 5]). Chagnon et al. (23) derived a significantly modular mycorrhizal network from the data of the mycorrhizal community described by Öpik et al. (18); they concluded that members of the genera Acaulospora and Scutellospora were mostly confined to a single module associated with “forest specialist plants,” whereas members of the genus Glomus were more generalist in their partner choice and were found mainly in the module comprising the common or “generalist” plant species.

Referring to our network, in the gypsum soil, where gypsophytes and gypsovags grow together, modules that interacted with gypsophytes were formed mostly by OTUs ascribed to the genus Glomus. However, the modules that interacted with gypsovags contained OTUs belonging to Glomus, Claroideoglomus, Paraglomus, Rhizophagus, Sclerocystis, Redeckera, and Diversispora. Interaction with a more phylogenetically diverse community of AMF can increase plant growth and potentially plant coexistence (19, 65, 66), mainly due to plant stress amelioration as a result of a high AMF functional complementarity in resource acquisition and delivery to the mutualistic association. Our results suggest that gypsophyte plant species, which might have specific adaptations for living in gypsum soils, are less dependent on the benefits of associating with a wide phylogenetic diversity of AMF. However, gypsovags, which presumably do not have specific adaptations to gypsum soils, might require the benefits of associating with a wide diversity of AMF to survive. This result opens new perspectives in our understanding of the influence of biotic interactions on plant community assembly rules, but future research under other ecological conditions will be required to ascertain whether this result reflects a general trend.

In conclusion, characterization of the ecological networks can be a valuable tool for ascertaining the potential influence of above- and below-ground biotic interactions (plant-AMF) on plant community composition. Our case study revealed that soil type can be a major factor shaping AMF communities, and there were some AMF groups with a tendency to interact differently with plants that had distinct ecological strategies (gypsophytes and gypsovags).