Effect of Rainfall-Induced Soil Geochemistry Dynamics on Grassland Soil Microbial Communities

Abstract

In Mediterranean-type grassland ecosystems, the timing of rainfall events controls biogeochemical cycles, as well as the phenology and productivity of plants and animals. Here, we investigate the effect of short-term (days) soil environmental conditions on microbial community structure and composition during a natural wetting and drying cycle. Soil samples were collected from a meadow in Northern California at four time points after the first two rainfall events of the rainy season. We used 16S rRNA microarrays (PhyloChip) to track changes in bacterial and archaeal community composition. Microbial communities at time points 1 and 3 were significantly different than communities at time points 2 and 4. Based on ordination analysis, the available carbon, soil moisture, and temperature explained most of the variation in community structure. For the first time, a complementary and more comprehensive approach using linear regression and generalized logical networks were used to identify linear and nonlinear associations among environmental variables and with the relative abundance of subfamilies. Changes in soil moisture and available carbon were correlated with the relative abundance of many phyla. Only the phylum Actinobacteria showed a lineage-specific relationship to soil moisture but not to carbon or nitrogen. The results indicate that the use of a high taxonomic rank in correlations with nutritional indicators might obscure divergent subfamily-level responses to environmental parameters. An important implication of this research is that there is short-term variation in microbial community composition driven in part by rainfall fluctuation that may not be evident in long-term studies with coarser time resolution.

INTRODUCTION

Soil is a limiting natural resource of great importance for human production of food, fiber, and fuel. It is highly biologically diverse and has the largest estimated biomass (51, 52). The activity of plant roots and soil microorganisms can considerably alter biogeochemical soil processes in the rhizosphere (26), which in turn can affect ecosystem function. Nutrients in soil are largely cycled from organic matter decomposition, but high carbon availability and nutrient limitation in the rhizosphere have been shown to select for bacterial communities with high weathering activity to replenish the nutrient pool (54). Rainfall, especially events involving wet-dry cycles, release labile carbon and nitrogen substrates into the soil through heterotrophic decomposition, microbial death, and cell lysis (7, 22, 55) and from soil organic matter bound in soil aggregates (15). However, over a growing season, the turnover of soil microorganisms has been demonstrated to be the largest source of dissolved organic and inorganic nitrogen available to plants (47). Soil microbial communities are thus central to soil nutrient cycling, both as drivers of nutrient release and as a source of nutrients. In the longer term, ecosystem-level models could be modified to include temporal and spatial response of microbial communities to environmental fluctuations, since these have important implications for carbon and nitrogen dynamics.

Mediterranean-type grassland ecosystems are highly dependent on the timing and the amount of rainfall (33), which controls plant productivity (32) and soil respiration (23). Variation in precipitation patterns will affect plant and animal phenology, food web structure, and nutrient cycling. We have previously shown that in Mediterranean-type grassland ecosystems, microbial communities responded to extreme climatic events, but these responses were short-lived and left little or no legacy (13). These microbial communities were shown to be robust to seasonal changes in precipitation patterns across the years despite significant changes in above ground plant and invertebrate community composition (50). The changing climatic conditions of these ecosystems, with dry and hot summers and wet and mild winters, have been proposed to select for microbial populations that are resilient to a wide range of environmental changes (58).

Even though microbial community structure is resilient to fluctuations in soil moisture across seasons and years, in the short-term (days) and especially after the rewetting of a dry soil, microbial communities quickly respond to changes in soil water potential and the associated pulses of soil carbon and nitrogen compounds (6, 10, 18, 48). Carbon and nitrogen pulses will trigger soil respiration and the activity of microbial populations. The quantity and quality of nutrients, such as carbon and nitrogen, have been shown to drive changes in microbial community structure is some ecosystems (2, 19). Soil moisture and pH are also known to be key environmental parameters to drive microbial community structure (1, 17, 36, 57). Given the complexity of soil systems, the high diversity of soil microorganisms, and gaps in our understanding of the physiology and ecology of most soil microbes, it is difficult to identify the environmental variables driving soil microbial community structure and composition.

The aim of the present study was to explore how short-term (days) rainfall dynamics affect a grassland soil microbial community structure and composition. We also address two questions of significant importance for the field of microbial ecology. First, can shifts in microbial community structure be explained by linear and nonlinear correlations between soil environmental conditions and abundance of taxa in the microbial community? Second, to what extent are environmental responses linked to higher taxonomic order? We repeatedly sampled the top 5 cm of the soil profile over 21 days after the onset of winter rains and used the 16S rRNA PhyloChip (8, 16) to identify and quantify the relative abundance of up to 9,000 different taxa. For the first time, we use a complementary statistical approach utilizing generalized logical networks (GLN) and linear regression to detect nonlinear (identified by GLN) and linear (identified by linear regression) interactions among environmental parameters and community composition tracked at the subfamily level. GLN provides the advantage of detecting nonlinear interactions that may otherwise be overlooked by linear regression. By using both models, we expect to detect the most significant associations and provide a more comprehensive understanding of possible interactions during natural wetting and drying events.

MATERIALS AND METHODS

Field site and soil sampling.

Soil sampling was performed within a defined plot (20 by 12 m) at the northern end of South meadow at the Angelo Coast Range Reserve in Mendocino Country, California (39°44′17.7″N, 123°37′48.8″W). Soils were deep and well drained with a sandy loam texture, derived from Cretaceous marine greywacke sandstones and mudstones of the Franciscan complex. The dominant soils orders in the Mendocino area are alfisols, inceptisols, and ultisols with a mixed mineralogy (5, 53). The grassland plant communities at this meadow comprise a mixture of annual grasses and annual and perennial forbs such as Bromushordeaceus, Bromusdiandrus, Airacaryophyllea, Madiagracilis, Trichostemmalanceolatum, Eschscholziacalifornica, and Rumexacetosella. At the time of sampling (end of summer to the beginning of the winter season), the plots had residual litter from the previous year and the forbs were only starting to germinate, making it difficult to accurately identify the plant species and measure plant biomass throughout the sampling period.

Sampling dates were determined in relation to current weather conditions to study the dynamics of short-term soil moisture cycles between precipitation events. We used the weather station at the Angelo Reserve's headquarters to track the local rainfall and air temperature (data were obtained at 15-min intervals). In 2007, the winter rains started on 9 October. By the time of the first sampling (22 October [T1]), the reserve had received 165 mm of precipitation (see Fig. 1). Between two sampling dates, 26 October (T2) and 8 November (T3), there was only 1.8 mm of precipitation. On 11 November, the rainfall resumed, with 17.4 mm of precipitation before the final sampling on 12 November (T4). The average air temperature during the sampling period was 8.6°C, with a maximum of 20.0°C (20 October) and a minimum of 0°C (at night). To account for diel cycles, sampling was always initiated at 2 p.m. PST and finished within 2 h. At each sampling occasion, five squares (20 by 20 cm) were randomly selected. Plants and litter were removed, and the soil temperature was recorded at a depth of 3 cm. Soil cores (three cores at T1 and T2 and two cores at T3 and T4) were collected within each square to a depth of 5 cm (4.45 cm in diameter). Soil samples were stored in plastic bags at 4°C and transported to the on-site laboratory for further processing. Plant litter was brought to the lab and dried at 80°C for 72 h before recoding its weight.

Fig 1.

(i) Daily rainfall (black lines), air temperature (gray dots), and dates of sampling T1 (22 October), T2 (26 October), T3 (8 November), and T4 (12 November). (ii to vii) Dynamics of soil moisture (SM) (ii), soil temperature (ST) (iii), extractable organic carbon (EOC) (iv), ammonium (NH₄⁺) (v), pH (vi), and effective cation exchange capacity (ECEC) (vii) in the top 5 cm of the soil across all four sampling points.

Sample preparation.

Replicate cores of each sample were manually homogenized in plastic bags at the Angelo Reserve laboratory within 8 h of collection. One sample from T2 was excluded from further analysis due to a mistake during sampling mixing. From each homogenized soil sample, five subsamples were separated and transported back to the laboratory. The amount of soil and storage conditions were the following: (i) for gravimetric soil moisture, 10 g was stored in preweighed foil containers at room temperature; (ii) for extraction of inorganic nitrogen, 20 g was stored in sampling cups with a 1:5 dilution (wt/vol) of 2 M KCl at 4°C; (iii) for extraction of organic carbon and nitrogen, 20 g was stored in plastic bags at 4°C; (iv) for chloroform fumigation, 20 g was stored in plastic bags at −20°C; and (v) for DNA extraction, 20 g was stored in plastic bags on dry ice during transport and then stored at −80°C. The remaining soil was stored at 4°C until it was returned to the laboratory, where a subsample (2 g) was taken for pH measurements. The remaining soil was air dried at room temperature and dry sieved to a <2-mm fraction. This fraction was later subsampled to measure exchangeable cations.

Environmental variables. (i) Gravimetric soil moisture.

Gravimetric soil moisture was calculated from the soil dry weight after oven drying 10 g of fresh soil at 105°C until a constant weight was achieved (minimum 24 h). Soil moisture was used to normalize nutrient measurements from moist soil extracts after subtracting the dry weight of the >2 mm fraction (collected by wet sieving and drying at 80°C for 72 h).

(ii) Extraction of inorganic nitrogen.

Within 24 h of sampling, inorganic nitrogen, NH₄⁺ and NO₃⁻, were extracted by shaking in 2 M KCl at 150 rpm for 1 h (40). Extracts were filtered through a Whatman no. 1 filter (prewashed with 2 M KCl) and stored at −20°C for further analysis. One blank was included for each set of extractions. At the end of the sampling period, all extracts were shipped to the UC Davis Agricultural and Natural Resources Analytical Laboratory (Davis, CA) for analysis of NH₄⁺ and NO₃⁻ concentrations using an automated flow injection analyzer method (27, 34). The amount of extracted NH₄⁺ and NO₃⁻ is given as μg of N per g of dry soil.

(iii) Extraction of organic carbon (EOC).

Within 24 h of sampling, organic carbon and nitrogen were extracted in 0.05 M K₂SO₄ (1:5 [wt/vol]) by shaking at 150 rpm for 2 h (56). Extracts were filtered by using a Whatman no. 1 filter (prewashed with 0.05 M K₂SO₄), and 50 ml was immediately used to measure extractable organic carbon in a 1010 TOC analyzer (OI Analytical, College Station, TX). Samples were diluted in Milli-Q water (1:2 or 1:5) to fit within the KHP standard range of 0 to 20 mg of C per ml. EOC is expressed as μg of C per g of dry soil. The remaining extract was frozen until the end of the sampling period.

(iv) Microbial biomass carbon and nitrogen.

Soil microbial biomass was estimated by fumigating 20 g of frozen soil for 7 days in a desiccator with an ethanol-free chloroform atmosphere. Fumigated samples were extracted with 0.05 M K₂SO₄ according to the procedure described above. Microbial biomass carbon was calculated as microbial biomass C = EC/kEC, where EC = (organic C extracted from fumigated soils) − (organic C extracted from nonfumigated soils) and kEC = 0.45 (3), and is expressed as μg of C per g of dry soil.

Extracts from fumigated and nonfumigated samples were digested in 1:1 (vol/vol) with 5% alkaline potassium persulfate reagent by autoclaving for 40 min at 121°C (9) to estimate the microbial nitrogen. Nitrate concentrations in the samples were measured with a QC8000 flow injection analyzer (Lachat Instruments, Milwaukee, WI). Microbial biomass nitrogen was calculated as microbial biomass N = EN/kEN, where EN = (total N extracted from fumigated soils) − (total N extracted from nonfumigated soils) and kEN = 0.54 (8), and is expressed as μg of N per g of dry soil. The C/N ratios were calculated as weight ratios.

(v) pH.

Within 24 h of sampling, 2 g of fresh soil was extracted in a 1:5 dilution (wt/vol) of 0.01 M CaCl₂ by shaking for 2 h at ∼100 rpm. The clear supernatant was transferred to clean tubes after centrifugation at 6,000 rpm for 5 min at 4°C, and the pH was measured (Fisher Acumet AR 20; Fisher Scientific).

(vi) Exchangeable cations.

Exchangeable cations were extracted from 4 g of air-dried soil (<2 mm) in a 1:5 (wt/vol) dilution of 0.1 M BaCl₂ by shaking on a vertical shaker at 45 rpm for 2 h (25). One internal extraction control and one blank were included with every set of extractions. The supernatant was separated by centrifugation at 6,000 rpm for 5 min at 4°C. Exchangeable cations Al, Ca, K, Mg, and Mn were analyzed by using the inductively coupled plasma with optical emission system (ICP-OES) with a Perkin-Elmer 5300 DV optimal emission ICP with autosampler. Blanks and internal extraction controls were included in the analysis. Quality controls were performed between every tenth sample, and scandium was included for normalization when measurements were drifting. Cations are expressed in μg per g (dry weight) of soil. The effective cation-exchange capacity (ECEC) was calculated as described by Hendershot and Lalande (25) using the following formulae: (i) M⁺ cmol (+)/kg = C cmol (+)/liter × (0.03 liter/g [dry weight] of soil) × 1,000 g/kg × DF, followed by (ii) ECEC cmol (+) kg⁻¹ = Σ M+ cmol (+) kg⁻¹.

Characterization of soil microbial communities. (i) Soil DNA extractions and amplification of bacterial and archaeal 16S rRNA genes.

DNA was extracted from 0.25 g (approximate dry weight) of homogenized soil using a power soil DNA kit (Mo Bio, Carlsbad, CA) according to the manufacturer's instructions. Approximately 6 to 10 ng of DNA was used to amplify the 16S rRNA genes from each of 19 samples. A temperature-gradient PCR was performed for each sample using the primers 27F (5′-GTTTGATCCTGGCTCAG-3′) and 1492R (5′-GGTTACCTTGTTACGACTT-3′). For the archaeal 16S rRNA gene amplification, two rounds of PCR were performed. For the first round, one reaction per sample was performed using the primers 4Fa (5′-TCCGGTTGATCCTGCCRG-3′) and 1492R, and for the second round a gradient PCR was performed using the primers 23F (5′-TGCAGAYCTGGTYGATYCTGCC-3′) and 1406R (5′-ACGGGCGGTGWGTRCAA-3′). The PCRs had a final volume of 25 μl containing a final concentration of 1× TaKaRa ExTaq PCR buffer with MgCl₂, a 300 pM concentration of primers, 1 μg of bovine serum albumin/μl, 200 μM deoxynucleoside triphosphates, 2.5 U of ExTaq DNA polymerase (TaKaRa Mirus Bio Inc., Madison, WI), and Milli-Q water to complete the volume. The PCR cycle for the bacterial 16S rRNA amplification was performed with an initial denaturation at 95°C for 3 min, followed by 25 cycles of 95°C for 30 s, annealing gradient from 48 to 60°C for 25 s, extension of 72°C for 2 min, and a final extension of 72°C for 10 min. The first round of archaeal PCR had a initial denaturation at 95°C for 3 min, followed by 25 cycles of 95°C for 30 s, annealing temperature of 50°C for 25 s, and extension of 72°C for 2 min and then a final extension of 72°C for 10 min, and the second round had the same cycling but with an annealing gradient of 53 to 65°C for 25 s. Amplicons from each sample and temperature were pooled and then purified using a QIAquick PCR purification kit (Qiagen, Maryland) and quantified using gel electrophoresis.

(ii) 16S rRNA microarrays.

From each pool of 16S rRNA genes, 500 ng of bacterial and 100 ng of archaeal DNA was fragmented, biotin labeled, and hybridized to a 16S rRNA gene Affymetrix microarray (PhyloChip) as described in detail elsewhere (8). PhyloChip washing, staining, and scanning were performed as described elsewhere (39). PhyloChip data were filtered and summarized at the subfamily level (ca. 94% sequence homology). A representative operational taxonomic unit (OTU) from each subfamily was selected based on the best probe set (gPM) and the greatest number of passing scores (a positive fraction of >0.90) across all samples. This procedure identified a total of 400 unique subfamilies that were further analyzed, 11 of which belonged to the Archaea. The PhyloChip data were normalized to the total array intensity to obtain the relative abundance of each taxon within the community. In the longer term, ecosystem-level models could be modified to include temporal and spatial response of microbial communities to environmental fluctuations, since these have important implications for carbon and nitrogen dynamics.

Data analysis. (i) Environmental variables across time.

One-way analysis of variance (ANOVA) using the standard least-square modeling, followed by a post hoc Tukey HSD test, was used to identify variables that significantly changed over time. All statistical analyses were performed using the JMP software (SAS Institute, Inc.).

(ii) Change in microbial community composition.

Nonmetric multidimensional scaling (NMS) (11, 35) and multiresponse permutation procedures (MRPP) (42, 43) were used to visualize and test dissimilarities in community composition based on the relative abundance of 400 unique subfamilies between sampling points. NMS was run in autopilot mode with 500 iterations (slow and thorough) and Sorensen (Bray-Curtis) distance measure. MRPP was also run with the Sorenson (Bray-Curtis) distance measure. Pearson and Kendall correlation coefficients between each environmental variable and the ordination axes were also obtained from the NMS analysis and sorted by decreasing r values. Ordination analysis was performed using PC-ORD version 5.26 (41). Taxa that significantly changed with time and the directionality of their response were identified using ANOVA and the post hoc Tukey test, respectively. Analyses were performed using JMP (SAS Institute, Inc.).

(iii) Associations among environmental parameters and between microbial taxa and the environment.

We detected associations among environmental variables and between the relative abundance of taxa and the measured environmental parameters by using two methods, GLN and linear regression. We attempted to capture nonlinear associations by GLN modeling and linear associations by linear regression. By using both models, we expect to detect most associations and provide a more comprehensive understanding of possible interactions.

GLN is a discrete-value and discrete-time dynamic system model. In a GLN, nodes represent variables and directed edges indicate associations from parent nodes (independent variables) to child nodes (dependent variables) (49). The association between parent and child variables is represented by a generalized truth table that maps parent values to child values. An association among variables at the same time point has no directionality.

We applied both GLN modeling and linear regression to examine the strength of associations between the relative abundance of taxa and the environmental variables observed at the same time point. First, we infer associations between environmental variables and taxa, represented as a GLN. Nodes in a GLN represent environmental variables and taxa abundance, and edges are associations between them. Before GLN modeling, values of relative abundance of taxa and the measurements of environmental variables were quantized to low, medium, and high levels by an optimal distance-based clustering algorithm implemented in the Ckmeans.1d.dp package in R (59). The associations and their significance are established between observed measurements of environment variables and relative abundance of taxa.

We also performed pre- and postprocessing. Two outliers in soil moisture measurements (likely measurement errors) were replaced by the median of soil moisture from all of the samples at the same time point. Although a taxon can be influenced by more than one environmental variable, our analysis only detected significant one-one associations because more-to-one associations became insignificant due to the limited sample size. All P values were further adjusted for multiple testing effects by the Benjamini-Hochberg method (4). Adjusted P values were used for the interpretation of the data and are presented here.

(iv) Environmental and lineage-specific relationships.

We used lineage-specific analysis in UniFrac (37) to determine the taxonomic level at which nutrient availability influenced the relative abundance of bacteria, using the correlation scores as environmental variables and the nonlinear associations from GLN analysis. The phylogenetic tree of the identified subfamilies was derived from the GreenGenes PhyloChip version G2 tree. Lineage-specific analyses were performed along the tree branch length in 0.01 to 0.02 increments, depending on the node density along the tree, using a G-test. The analysis was terminated when a lineage had fewer than four descendants. Lineages significantly associated with either positive or negative correlations to environmental variables are highlighted in the phylogenetic tree (see Fig. 5).

Fig 5.

Phylogenetic tree of 400 analyzed subfamilies, derived from the GreenGenes PhyloChip version G2 tree. Branches with no significant linkage to environmental variables are collapsed to the taxonomic levels phyla or order depending on the species represented by the branch. Branches representing five or more subfamilies are labeled, with the number of subfamilies given in parentheses. Nodes with significant lineage-specific responses to soil moisture are indicated with an asterisk (*, P < 0.001) for the Actinobacteria (in the framed box) and the Actinomycetales (in the gray-shaded box). The correlation in abundance of each subfamily within the Actinobacteria in relation to SM is illustrated by “■” for negative correlation, “□” for positive correlation, and “▩” for no correlation. The scale bar indicates the substitutions per site.

RESULTS

Environmental variables dynamics.

The soil environmental variables that changed significantly with time were soil moisture (SM), soil temperature (ST), pH, extractable organic carbon (EOC), ammonium (NH₄⁺), and effective cation exchange capacity (ECEC) (Fig. 1; see also Table S1 in the supplemental material). No significant changes were observed for Al, Ca, K, Mg, and Mn, for microbial biomass carbon, nitrogen, and C/N ratios, and for above-ground litter biomass (see Table S1 in the supplemental material). As expected, SM decreased with time after rainfall ceased and was higher during the second rainfall (T4) than at the first sampling (T1), which occurred 3 days after the first major rain of the season. Decreasing air temperature resulted in decreasing ST during the sampling period, with the lowest soil temperature at T4. The EOC was unchanged across all time points with the exception of T3 when it was significantly higher. Soil available NH₄⁺ decreased with time, reaching its lowest value at T3. Soil pH and ECEC changed significantly between T2 and T3, where soil pH decreased and ECEC increased (Fig. 1; see Table S1 in the supplemental material).

GLN analysis identified statistically significant associations among SM, ST, pH, ECEC, and EOC at the same time point (Fig. 2). Soil moisture was found to have a strong (P ≤ 0.05) association with EOC and soil temperature. Soil temperature is also found to have a strong association (P ≤ 0.05) with ECEC. The linear regression model also detected several statistically significantly (P ≤ 0.05) associations among environmental variables. Specifically, soil moisture is negatively associated with EOC and ST, EOC is negatively associated with pH, and ST is negatively associated with ECEC. Figure 2 and Table S3 in the supplemental material show these significant associations detected by both GLN and linear regression.

Fig 2.

Schematic representation illustrating the associations among environmental variables (dotted lines) according to generalized logical network (GLN, *) and linear regression (LR, ○) and the percentage of the variation explained by each environmental variables (solid lines) on changes in soil microbial community structure at each time point based on Pearson and Kendall correlations from the nonmetric multidimensional scaling analysis (NMS). Stress value = 4.65. See the NMS statistics in Table S2 in the supplemental material.

Change in soil bacterial and archaeal community structure.

NMS demonstrates significant differences (MRPP P < 0.0001) in the soil bacterial and archaeal communities among the four time points (Fig. 2). Two main clusters were observed: one consisting of T1 and T3 and the second consisting of T2 and T4. Most of the variation is explained on axis 1 (99%, Fig. 2). The Pearson and Kendall coefficient (r values), obtained from the NMS analysis, suggests that soil moisture and pH have a negative correlation with the observed shift in microbial community structure, whereas EOC, soil temperature, and NH₄⁺ showed a positive correlation (see Table S2 in the supplemental material, R2 > 0.1).

The observed separation in the NMS ordination is in agreement with a significantly higher relative abundance (ANOVA; P < 0.05) of members of the phyla Acidobacteria, Firmicutes, Proteobacteria, Bacteroidetes, Chloroflexi, and Verrucomicrobia at time points T2 and T4, whereas members of the phyla Actinobacteria and Cyanobacteria had higher relative abundance at time points T1 and T3 (Fig. 3).

Fig 3.

Percentages of subfamilies within each phylum that showed a significantly (P < 0.05 [ANOVA]) higher relative abundance at T1 and T3 (black bars), T2 and T4 (dark gray bars), and no response (light gray bars). The total numbers of subfamilies within each phylum that were identified by PhyloChip are shown in parentheses.

Associations between relative taxon abundance and environmental variables.

We applied both GLN and linear regression modeling to identify associations between environmental variables and relative abundance of taxa. Using GLN modeling, soil moisture and soil temperature were found to be the main factors associated with the relative abundance of taxa: 136 (34%) and 79 (20%) taxa are associated with soil moisture and soil temperature, respectively (see Table S3 in the supplemental material). No significant associations were observed for EOC, NH₄⁺, pH, and ECEC. In Fig. S1 in the supplemental material we present scatter plots of the two most significant nonlinear associations (detected by GLN) between the relative abundance of representative OTU and environmental variables (soil moisture and soil temperature). Using linear regression, 303 (76%) taxa were associated with soil moisture, 74 (19%) with EOC and 1 (0.25%) with pH (see Table S3 in the supplemental material). No other environmental variable showed significant associations after Benjamini-Hochberg correction. Figure S2 in the supplemental material shows the scatter plots of the most significant linear associations, detected by linear regression, between relative taxa abundance and environmental variables. If we combine linear regression and GLN results, 74 (19%) taxa are associated with EOC, 1 (0.25%) with pH, 305 (76%) with soil moisture, and 79 (20%) with soil temperature (see Table S3 in the supplemental material). Statistically significant associations between environmental variables and the relative abundance of taxa were used to evaluate the significance of these observations at a higher taxonomic rank.

Relationship between phylogeny and environment variables.

From all of the environmental variables measured, EOC and soil moisture showed significant (adjusted P < 0.05) positive or negative correlations with the relative abundance of many subfamilies within the Actinobacteria, Acidobacteria, Proteobacteria, Bacteroidetes, Firmicutes, Cyanobacteria, Verrucomicrobia, and others (Fig. 4). The majority of subfamilies that correlated significantly with EOC showed a negative association with this environmental variable. On the other hand, a higher percentage of subfamilies within most phyla (with the exception of the Cyanobacteria and Actinobacteria) showed a positive correlation with soil moisture. This suggests that changes in the relative abundance of these phyla are associated with changes in soil moisture. However, the phylum Actinobacteria was the only clade to show a lineage-specific relationship with soil moisture (Fig. 5). For all other lineages, the response was not consistent at any taxonomic level tested. Nonlinear associations from GLN revealed no lineage-specific relationship.

Fig 4.

Percentages of subfamilies within each phylum that showed a positive (gray bars) or negative (black bars) correlation (P < 0.05 [Pearson]) with EOC (top panel) and SM (bottom panel). The total numbers of subfamilies within each phylum that were identified by PhyloChip are shown in parenthesis.

DISCUSSION

Over a 3-week period in a Mediterranean-type grassland ecosystem, the soil microbial community composition changed in response to fluctuations in soil moisture, temperature, pH, and carbon and ammonium availability during a natural drying and rewetting cycle following the first rains of the winter season. For the first time, GLN was used to analyze associations among soil chemical and abiotic variables. We demonstrate that short-term (days) fluctuations in soil moisture and temperature are associated with significant changes in extractable organic carbon and ECEC and that these variables, as well as ammonium and pH, influence soil microbial community structure (Fig. 2).

After a long dry summer, the first rainfalls of the winter season will change the soil water potential triggering plant germination, litter decomposition, and microbial activity. Labile carbon and nitrogen sources are released and recirculated in the ecosystem by active microbiota (6, 28, 31, 38). The active microbial communities and the growing plants quickly assimilate nitrogen (21, 24, 29, 30, 46), possibly explaining the rapid decrease in NH₄⁺ availability following the first rainfall event (T1 to T3, Fig. 1). Subsequent nitrogen limitation may induce further microbial decomposition of the organic C-N pools, resulting in a net release of organic carbon (10), explaining the concomitant extractable organic carbon pulse at T3 (Fig. 1).

NMS analysis demonstrates a clear separation in microbial community composition between time points 1 and 3 compared to time point 2 and 4. The relative abundance of many subfamilies within the Acidobacteria, Verrucomicrobia, Bacteroidetes, Firmicutes, and Proteobacteria was higher at time points 2 and 4 and showed a significant positive correlation with soil moisture and ammonium (unadjusted P < 0.05) and a negative correlation with EOC (Fig. 3 and 4). Earlier studies have demonstrated contrasting responses to nutrient availability among these phyla. Based on their high abundance in soils with low carbon availability, the Acidobacteria have previously been classified as oligotrophic (17), which is in agreement with our observations. However, contrary to previous studies (14, 17), the Bacteroidetes and Proteobacteria showed a similar response to low carbon availability in our study. Due to the complexity of changing nutrient conditions in natural soils, single variables such as carbon availability may not be sufficient to determine ecological strategies among soil microorganisms. Furthermore, our results demonstrate that not every member of each phylum has the same responses to environmental variables.

In the present study, Actinobacteria was the only phylum that showed a lineage-specific response, and this was only seen in response to SM (Fig. 5). One sublineage (the Actinomycetales clade in Fig. 5) shows a consistent negative correlation, whereas other sublineages display a positive correlation. This behavior, where ecological and phylogenetic clusters match, has been previously described as ecological coherence (17, 44). Our results suggest that ecological coherence at a higher taxonomic rank is more an exception than a universal rule for soil microbial ecology. Increased soil moisture can create anaerobic microniches detrimental to strict aerobic organisms such as most Actinobacteria but suitable for strict or facultative anaerobes. The subsets of Actinobacteria that are facultative or obligate anaerobes are probably those that responded positively to increased SM (20).

Many prior studies of ecological strategies among soil microorganisms and nutrient availability have analyzed wide phylogenetic groups (phylum or class level) using phospholipid fatty acid analysis (1, 12, 19) or quantitative PCR with phylum-specific primers (17, 45). The results obtained from these studies may potentially be driven by the relative abundance of a subset of dominating taxa. In the present study, 16S rRNA gene PhyloChip analyses provided relative abundance data at the subfamily level, and the responses of different phylogenetic groups are derived from the responses of all detected subfamilies within the group. We acknowledge that our ability to detect responses to environmental variables is limited by the number of taxa represented on the microarray assay. However, previous studies have shown that array results are comparable to clone libraries (16).

In conclusion, short-term changes in moisture and nutritional and geochemical conditions during a natural rainfall cycle had a significant effect on the structure of soil microbial communities. Soil moisture, temperature, and available carbon and nitrogen were correlated with changes in community composition. Because of the complexity in the interaction among soil chemical, physical, and nutritional properties, it remains difficult to identify the main soil variables driving changes in the relative abundance of many phylogenetic groups. Here, we show the advantage of using complementary methods that analyze linear (linear regression) and nonlinear (GLN) association in order to identify a broader range of interactions between environment variables and microbial abundance. For instance, the association between soil temperature and extractable organic carbon was only identified by GLN, whereas the association between extractable organic carbon and pH was only identified by linear regression. Similarly, GLN showed significant associations between some taxa and soil moisture and temperature that were not identified by linear regression (see Fig. S1 in the supplemental material). Our findings also emphasize the importance of fine-scale temporal resolution in understanding soil microbial community dynamics. The findings presented here complement our prior research, in which we showed year-to-year resilience in overall microbial community composition despite differences in imposed soil moisture conditions (13). Taken together, this research emphasizes that soil microbial communities respond rapidly to changes in nutrient availability and other conditions following major seasonal events (such as rainfall following a prolonged drought) but maintain their overall structure over the longer term.