Functional redundancy in local spatial scale microbial communities suggests stochastic processes at an urban wilderness preserve in Austin, TX, USA

ABSTRACT

Empirical evidence supports selection of soil microbial communities by edaphic properties across large spatial scales; however, less is known at smaller spatial scales. The goal of this research was to evaluate relationships between ecosystem characteristics and bacterial community structure/function at broad taxonomic resolutions in soils across small spatial scales. We employed 16S rRNA gene sequencing, community-level physiological profiling and soil chemical analysis to address this goal. We found weak relationships between gradients in soil characteristics and community structure/function. Specific operational taxonomic units did not respond to edaphic variation, but Acidobacteria, Bacteroidetes and Nitrospirae shifted their relative abundances. High metabolic diversity within the bacterial communities was observed despite general preference of Tween 40/80. Carbon metabolism patterns suggest dominance of functional specialists at our times of measurement. Pairwise comparison of carbon metabolism patterns indicates high levels of functional redundancy. Lastly, at broad taxonomic scales, community structure and function weakly covary with edaphic properties. This evidence suggests that stochasticity or unmeasured environmental gradients may be influential in bacterial community assembly in soils at small spatial scales.

INTRODUCTION

Soil microbial communities drive food webs by contributing to the organic matter pool through their biomass (Schlesinger and Andrews 2000) acting as a food source (Scheu 2002), and mediating decomposition and nutrient cycling (Lladó, López-Mondéjar and Baldrian 2017). As a result, soil microbial communities significantly affect aboveground communities through plant growth and plant diversity (Van Der Heijden, Bardgett and Van Straalen 2008; Schnitzer et al. 2011; Wagg et al. 2014) and are relevant under global change scenarios. Abiotic selection strongly affects the structure and function of soil microbial communities, summarized by Baas Becking's tenet ‘everything is everywhere, but the environment selects’ (De Wit and Bouvier 2006). Bulk soil characteristics [pH, soil organic matter (SOM) and moisture] are essential for microbial functions, including energy production (e.g. by changing potential electron acceptors), biomass (Fontaine, Mariotti and Abbadie 2003) and physiology (Reed 1990). The extent to which subtle variation in soil characteristics influences microbial community structure and function remains poorly described at small/local spatial scales (tens to hundreds of meters).

At large spatial scales (>100 km), studies have found differences in microbial community composition and/or diversity across environmental gradients (Fierer and Jackson 2006; Lauber et al. 2009; Griffiths et al. 2011; Garcia-Pichel et al. 2013; Xue et al. 2018) where gradients are more pronounced. Previous research has demonstrated that community structure shifts across continental soil pH gradients (Lauber et al. 2009; Griffiths et al. 2011; Docherty et al. 2015; Xue et al. 2018), and that relative abundances of specific bacterial taxa at broad taxonomic resolutions (e.g. Acidobacteria, Bacteroidetes and Proteobacteria) respond strongly to differences in soil pH (Jones et al. 2009; Lauber et al. 2009; Rousk et al. 2010). pH may be an important control on community structure and function at large scales, but this may be attenuated at local scales due to subtler variations in soil chemistry. Likewise, effects may be clade specific. For example, arbuscular mycorrhizal fungi undergo shifts in community structure due to gradients in pH across short distances (Dumbrell et al. 2010).

While community structure and function covary with specific environmental parameters at large scales, variability in soil characteristics at small scales is less pronounced. Functional redundancy may be responsible for this variability (Allison and Martiny 2008; Rousk, Brookes and Bååth 2009). Removal of a taxon may not drastically alter ecosystem function as another taxon with the same functional potential fills the open niche allowing the ecosystem to enter an alternative stable state (Shade et al. 2012). Stochastic processes also influence the structure and function of microbial communities (Caruso et al. 2011; Stegen et al. 2012). For example, community assembly is shaped by neutral ecological processes following significant ecological disturbances such as wildfires (Ferrenberg et al. 2013). Dispersal, drift and facilitation also influence community assembly (Emerson and Gillespie 2008; Kembel 2009). Community structure is likely a combination of deterministic and stochastic processes, although further research on the ecological conditions that promote stochasticity is needed (Caruso et al. 2011). Characterizing patterns of microbial community structure may elucidate which ecological processes control small-scale patterns of biodiversity.

The goal of this research was to investigate whether gradients in edaphic properties over local spatial scales strongly influence bacterial community structure/function at broad taxonomic resolutions and mirror patterns found at larger spatial scales. We hypothesized that if communities are assembled largely through deterministic mechanisms, we will observe a strong correlation between edaphic variation and community structure/function. Specifically, we expect that due to the small size of microbes and the high surface area/volume of soil, pH, moisture and organic matter content would drive local-scale microbial community structure and function and mirror patterns found at larger scales. A combination of bulk and chemical soil characterization, analysis of community-level physiological profiles and 16S ribosomal RNA gene sequencing was used to test these hypotheses. This research adds novel evidence for understanding of the local processes that govern the links between microbial community structure, function and assembly within soil ecosystems.

MATERIALS AND METHODS

Site description

Wild Basin Wilderness Preserve is a 227-acre urban preserve located in Austin, TX, USA (30.3103455°N, −97.8234956°W), on the border between a subtropical humid climate and a subtropical subhumid climate. The preserve is characterized by large deposits of limestone with numerous Ashe Juniper, Red Oak and Live Oak trees. Wild Basin is affected by the rapid urbanization occurring in and around Austin (USCB 2010). The preserve is bordered by Texas State Highway Loop 360 to the west, which is a corridor of active suburban development. In 1961, a high-intensity wildfire burned ∼60% of the woody vegetation across the preserve (Westlake Fire Department, pers. comm.). Thirteen of 19 sites in this study were located within the burned area. Since 1996, sites sampled in this study have hosted established woody plant communities.

Soil sampling

Soils were collected at 1–2-month intervals between May 2016 and June 2017 from 20 long-term monitoring sites (Fig. 1; Table S1, Supporting Information). Specific sampling days were chosen when there had been no precipitation that day and for the preceding 7 days. At each site, three samples were collected within a 5 m² plot. Samples were collected in triplicate at ∼8.0 cm depth using a sterilized soil core. Triplicates at each site were homogenized before analysis and stored at −20°C for 24 h before use. Methods used for characterizing edaphic properties can be found in the Supporting Information.

Figure 1.

Map of Wild Basin Wilderness Preserve showing site locations and elevation, generated using a USGS DEM and Esri ArcMap (10.4.1).

Microbial community function

Community-level physiological profiling (CLPP) was measured using BiOLOG EcoPlates (BiOLOG, Hayward, California, USA). EcoPlates are 96-well plates consisting of different carbon guilds (carbohydrates, polymers, carboxylic acids, amines/amides, amino acids) that allow measurement of metabolism through the reduction of a tetrazolium dye. Thirty-two carbon compounds are present with one water control. These are found in triplicate on each plate for replication. Soil (10.0 g in 1.0 L sterile 1% saline DI water solution) was placed in a shaking incubator for 3 h at 21°C. EcoPlates were inoculated with 100.0 μL of 10⁻³ diluted soil from this initial solution and placed (covered with a sterile and closed lid) in a Multiskan FC plate reader at 21°C (Fisher Scientific, USA) for 72 h at OD595 nm (Gomez, Ferreras and Toresani 2006; Gryta, Frąc and Oszust 2014). It should be noted that EcoPlates have biases based on how long it takes a well to reach exponential growth phase as compared with all other wells (Garland 1997). Means of the triplicate absorbance values for each substrate from each plate can be found in Table S2 (Supporting Information), which are used in Fig. 2B and for calculating average well color development (AWCD).

Figure 2.

(A) Bar plot of carbon substrate metabolism separated by carbon guild. (B) Heat map of hierarchically clustered AWCD values separated by site with carbon substrate on the left and carbon guild on the right. (C) Graph: Spearman correlation between functional diversity and taxonomic diversity with correlation coefficient and P-value in the bottom left. Heat map: Shannon taxonomic and functional diversity values separated by site. NA values are due to errors with sequencing. (D) Bar plot of operational taxonomic unit (OTU) occupancy separated by proportion of sites occupied by an individual OTU. (E) Heat map of random forest regression variable importance where lighter colors indicate greater importance to the model. To the right are the correlation coefficient and root mean standard error (RMSE in AWCD values).

Microbial community structure

DNA from samples was isolated using the Mo Bio PowerSoil DNA Isolation Kit and DNA concentrations were measured using a NanoDrop Spectrophotometer. Sequencing was conducted at MR DNA (Shallowater, TX, USA) on an Ion Torrent PGM following the manufacturer's guidelines. Subsequent sequence analysis was conducted with an in-house proprietary pipeline. To summarize, primer sequences 515F (5′-GTGYCAGCMGCCGCGGTAA-3′)–806R (5′-GGACTACNVGGGTWTCTAAT-3′) (Earth Microbiome Project 2016) barcodes, and reads <150 bp were removed; sequences with ambiguous base calls and with homopolymer runs exceeding 6 bp were also removed. Sequences were denoised and chimeras removed. Operational taxonomic units (OTUs) were defined by clustering at 97% similarity. OTUs were taxonomically classified using BLASTn against the Ribosomal Database Project (RDP-II) reference database. We obtained 201 304 raw reads with an average length of 270 bp and 106 092 reads after quality control, rarefied to the minimum number of sequences across sample (4982) that were clustered into 832 OTUs and relative abundances were calculated. Sequences have been deposited in the Sequence Read Archive (SRA) under SUB7125886 and OTUs are in Table S6 (Supporting Information).

Data analysis

Statistics were calculated in R (3.6.0) using the vegan (Oksanen et al. 2013) and graphics with ggplot2 (Wickham et al. 2008). All variables were examined for normality using Shapiro–Wilk tests. Outliers were identified as 1.5 times the interquartile range and removed. Non-normal data were log transformed to normality.

CLPP data were normalized by calculating AWCD (Average Well Color Development) that was calculated as (OD: Optical Density) ΣOD_i/n_(substrates) at 72 hours for each EcoPlate as a metric for total CLPP activity (Gomez, Ferreras and Toresani 2006) that takes into account differences in inoculum density (Garland 1997) and would capture metabolic activity that molecular methods may miss. This equation was modified to ΣOD_{(triplicate substrates)}/n_{(all substrates)}. Hierarchical clustering using Bray–Curtis dissimilarity on mean values of each substrate per plate was done to determine similarity in substrate use. Microbial and functional diversity were calculated using the Shannon index (Shannon and Weaver 1949). Edaphic, taxonomic and functional dissimilarities were measured using the Bray–Curtis metric with correlation and significance with Mantel tests and PERMANOVAs (Permutational Analysis of Variance) (n = 999).

Mann–Whitney tests with Bonferroni correction identified differences in community function (Table S3, Supporting Information). Random Forest regression with the caret package (Kuhn 2008) modeled community function and edaphic variation. AWCD values for the 10 most used substrates with pH, SOM, moisture, phosphorus and nitrate concentrations were used in the models. Data were scaled and centered with 25 bootstrapped resamples conducted to test accuracy and root mean standard error was used to evaluate error. The varImp function was used to evaluate variable importance.

OTU counts were transformed into relative abundances for analysis. Bonferroni-corrected Mann–Whitney tests identified pairwise differences in community composition (Table S4, Supporting Information). Spatial autocorrelation was tested between OTU level and CLPP Bray–Curtis distance and Euclidean geographic distance matrices (Sokal and Wartenberg 1983) with a Mantel test. OTU-level associations between environmental variables were assessed using Bonferroni-corrected Spearman correlations. Occupancy was measured by counting the presence of an OTU at a site and taking the sum of all sites. These values were then ranked and broken into four equal categories. Overall, community structure was explored using non-metric dimensional scaling (Bray–Curtis distance) and fitting of edaphic properties with function envfit(n = 999).

RESULTS

Edaphic properties

Soils were slightly alkaline, with average pH of 7.64 ± 0.27. The range in soil pH was 7.18–8.26 with only two sites having a pH above 8.0 (sites 7.1, 9.1). Soil moisture was more variable, with an average value of 17.83 ± 6.0% and a range of 7.91–30.71% across all sites. SOM had an average value of 12.51 ± 3.76%. No soil characteristics were significantly correlated with soil temperature, depth or elevation. Further properties are in Table S5 (Supporting Information).

CLPP

Total metabolic activity in the soil samples, as measured through AWCD (dimensionless units), ranged from 0.200 to 0.853. Polymers were the most used carbon guild followed by amines/amides, carbohydrates, amino acids and lastly carboxylic and acetic acids (Fig. 2A). A high degree of variability in substrate use was found in all compounds except d-xylose and 2-hydroxybenzoic acid that were used the least overall (Fig. 2B). The most used compounds (Tween 80 and Tween 40 used at 9/15 sites) were complex polymers made of ethylene oxide and the sugar sorbitol and demonstrated a wide range in use spatially (Fig. 2B).

A wide breadth of functional diversity was observed (Fig. 2C) that varied spatially with a range of 2.10–3.19 (Shannon index) with a mean of 2.87. Site 4.1 had the greatest diversity of carbon substrate use of all the sites, while site 3.1 had the lowest carbon substrate use diversity. Pairwise comparisons (Mann–Whitney tests) of AWCD across all compounds and sites revealed significantly different functional profiles of carbon substrate usage for 3 of the 19 sites (Table S2, Supporting Information). A Mantel test between Bray–Curtis dissimilarity matrices of CLPP data and edaphic properties showed an insignificant relationship between CLPP profiles and soil characteristics (r² = 0.02, P > 0.05).

Random Forest regression demonstrated that of the 10 most used carbon substrates, weak to moderate correlations (Fig. 2E; 0.260–0.606) were found between substrate metabolism and variation in edaphic properties. Overall, pH was the most important variable for modeling substrate metabolism followed by phosphorus, moisture, SOM and nitrate. No discernible pattern by carbon guild was observed.

Microbial communities

Proteobacteria comprised ∼33% of all reads with Actinobacteria being ∼30%. All measured phyla were present at all sites; however, their relative abundances differed. OTU-level Shannon diversity ranged from 4.42 to 4.84 with a mean of 4.56. Overall, community structure was stable with the exception of significant differences at sites 4.1, 12.1 and 12.2 (Table S3, Supporting Information). Functional diversity did not scale with taxonomic diversity (Fig. 2C; Rho = 0.13, P = 0.626). OTUs were either largely found at all sites (Fig. 2D; 31.08%) or had low occupancy across the preserve (45.77%). Community structure did not vary seasonally over the sampling period (PERMANOVA, r² = 0.19, P = 0.421).

A Mantel test demonstrated insignificant relationships between the OTU-level community composition and soil characteristics (r = 0.34, P > 0.05) or spatial autocorrelation (r = −0.12, P > 0.05). While individual OTUs did not significantly respond to variation in edaphic properties, relative abundances of Acidobacteria and Nitrospirae increased with soil moisture content (Rho = 0.63, P < 0.01) and relative abundances of Nitrospirae decreased across the SOM gradient (Rho = −0.69, P = 0.003). Likewise, relative abundances of Bacteroidetes (Rho = −0.49, P < 0.05) and Acidobacteria (Rho = −0.52, P < 0.05) decreased with increasing pH. Overall, community structure when observing all phyla did not respond to environmental variables (Figure S1, Supporting Information; envfit P > 0.05 for all edaphic properties).

DISCUSSION

The goal of this research was to test the relationship between bulk soil characteristics (pH, moisture and SOM), and the structure/function of local microbial communities at wide taxonomic resolutions. Mantel tests revealed insignificant correlations between edaphic properties and community structure or function, despite subtle variation in soil properties across sites. We did not find evidence for significant differences between communities based on taxonomic structure; however, specific taxa (Acidobacteria,Bacteroidetes,Nitrospirae) are significantly correlated to specific edaphic properties. Functionally, a wide range of carbon substrates were metabolized at all sites and the explanatory power of edaphic properties to carbon metabolism varied. Furthermore, communities were functionally redundant at the majority of sites despite significant differences in community composition at all but three sites.

When taken together, our results suggest local-scale microbial communities may assemble differently than they do at larger scales. Weak relationships between three phyla-level taxa and edaphic properties as well as multiple carbon metabolism patterns suggest that deterministic processes are not as strong at small spatial scales. Instead, local-scale stochastic processes may have greater relative importance in structuring microbial communities. Likewise, the lack of a relationship between taxonomic and functional diversity also suggests that the community is not deterministically assembled based on selection on ability to metabolize specific carbon compounds.

The relationship between environmental characteristics and microbial community structure

Microbial community structure as a whole was not determined by subtle variation in edaphic properties. Previous research has established soil pH as an important factor in the structure of microbial communities at large (Fierer and Jackson 2006; Hartman et al. 2008; Lauber et al. 2009). However, we did not find evidence that pH significantly structured soil microbial communities, nor a relationship between pH and microbial community diversity. This may be attributed to the buffering effects of carbonate and high calcium (Tavakkoli et al. 2015) in the underlying limestone bedrock of the Edwards Plateau (Barker, Bush and Baker 1994). Further, pH has less of an effect on community diversity in soils with pH higher than 7.0 (Fierer, Bradford and Jackson 2007), and the small spatial scales sampled here may not have sufficient variation in pH for significant environmental filtering of the community.

Despite weak selection by soil characteristics on the entire community, responses by individual taxa to pH were observed. Our finding of Bacteroidetes abundance having a negative response to soil pH contrasts with current literature (Lauber et al. 2009; Rousk et al. 2010). These studies take place at the continental scale (with stronger spatial gradients in the concentrations of electron donors and acceptors than in our study) and in laboratory manipulations and may not take into account responses of Bacteroidetes abundance along attenuated pH gradients.

Our findings complement literature identifying negative correlations of Acidobacteria and positive correlations of Nitrospirae with pH (Fierer, Bradford and Jackson 2007; Lauber et al. 2008; Jones et al. 2009). These results suggest that pH and soil moisture have greater influence for specific taxa, such as Acidobacteria. For others (e.g. Bacteroidetes), pH may be a proxy for other co-variables that have greater impact on a taxon's ability to grow, compete and proliferate in the soil environment. Furthermore, relative abundances of Acidobacteria and Nitrospirae increased with soil moisture content, while the relative abundance of Nitrospirae decreased across the SOM gradient. Soil moisture and SOM previously affected the structure and function of microbial communities in microcosms and across a climatic gradient (Drenovsky et al. 2010; Brockett, Prescott and Grayston 2012). Aside from measured edaphic properties, the roles of plant community composition (Myers et al. 2001; Yang et al. 2014; Reese et al. 2018; Sielaff et al. 2018) and soil texture (Bach et al. 2010) have been identified as factors in the structure of microbial communities, but those factors were unmeasured in this study.

CLPP profiles and edaphic properties

Our finding that community structure did not define how its members will function conflicts with other studies (Griffiths et al. 1998; Fierer et al. 2012). This may be attributed to functional redundancy, stochastic processes, insufficient variation of edaphic properties to establish functional niches or unmeasured variables such as competitive exclusion or symbiosis. Likewise, we measure community function at a point where all substrates are used and the community may be in late state succession dominated by specialists. Further analyses should investigate function at multiple time points and directly quantify inoculum density. Furthermore, this study only addresses carbon metabolism and other microbial functions have not been measured. Although all carbon substrates were used by every community in our study, there was a great deal of variation in terms of carbon substrate preference, which indicates a wide range of metabolic diversity. Many previous studies investigating the relationship between taxonomic and functional diversity took place at larger spatial scales where there were larger gradients in environmental characteristics. Previous research has found that enzymatic function in soils covaries with spatial scale (ranging from soil horizon to landscape) (Parsons, Smith and Murray 1991; Decker, Boerner and Morris 1999; Šnajdr et al. 2008). It is possible that despite the wide range in both bulk soil characteristics and chemistry the variation in an edaphic property (e.g. pH) was not a strong enough signal to influence a shift in taxon abundance or carbon metabolism at this geographic scale. It should be noted that variation in community growth/succession and metabolic rates at time of measurement remains unquantified and may influence measured function.

Discrete patterns in carbon substrate preference were observed as well as each substrate being utilized at every site, although to varying degrees of use. In agreement with literature (Reese and Maguire 1969; San Miguel, Dulinski and Tate 2007; Gryta, Frąc and Oszust 2014), we found that polymers Tween 40 and Tween 80 were extensively used by the community; however, the amount to which these polymers were metabolized varied spatially. These compounds are water-soluble surfactants consisting of partial fatty acid esters combined with ethoxylated sorbitan that may be a preferred substrate after easily degradable compounds are exhausted. This suggests that specialists at our measured time point are dominating the communities since Tween 40 and Tween 80 are the average most used substrates. Furthermore, this suggests the possibility of extracellular enzyme digestion (Skujiņscaron and Burns 1976; McShan et al. 2016) before sequestration by microorganisms given its large size and the energetic preference for diffusion of small molecules across cell membranes (Reese and Maguire 1969; Decad and Nikaido 1976).

In contrast, with significant differences in community structure, our results are suggestive of functional redundancy with regard to carbon substrate metabolism in that only 3 out of 19 sites had significantly different functional profiles. This structure–function disconnect offers two possible interpretations. First, taxa are equal in function due to weak deterministic structuring of communities, as supported by our finding that few taxa responded to variation in edaphic properties. Weak correlations between edaphic variation and carbon metabolism also support this idea. Second, the individual taxa may not be redundant but when combined with all the other microorganisms in the community result in the same function at the community level (Allison and Martiny 2008).

Potential for stochastic processes driving community assembly and function

With similar soil characteristics across the site, it is likely that abiotic selection does not exert a strong influence on microbial community structure and function. In contrast to community-wide patterns, individual taxa more sensitive to specific environmental conditions may be widely dispersed but their abundances would be more closely related to shifts in key edaphic properties. Our findings suggest that community assembly may be driven by factors other than selection alone, possibly stochastic processes such as dispersal and drift, as proposed by Nemergut et al. (2013), or traits such as genome size and functional potential (Barberán et al. 2012). The influence of latent variables such as interspecies interactions and aboveground plant feedback remains unquantified in this study. Other studies demonstrate disconnects between microbial community structure, function and environmental conditions (Ramette and Tiedje 2007; Mao et al. 2008; Graham et al. 2016). There is also evidence that certain environmental conditions, such as a high C:N in soils, and ecological disturbances like wildfire are associated with greater stochasticity in microbial community assembly (Nemergut et al. 2013; Evans, Martiny and Allison 2017). In fact, at our study sites, a wildfire in 1960 significantly altered the plant communities, and possibly the soil properties. This also suggests that understanding the legacy of global change effects can give insight into community structure (Meisner et al. 2018).

Recent research has elucidated stochastic processes that influence microbial community assembly, including historical contingency, dispersal, ecological drift and selection (Allison and Martiny 2008; Dumbrell et al. 2010; Vellend 2010; McMahon, Wallenstein and Schimel 2011; Evans, Martiny and Allison 2017), may be occurring at our study sites. Furthermore, the assembly of communities in which functional specialists dominate, such as in our study, may not be affected as strongly by environmental selection pressures (Langenheder and Székely 2011). This may be due to a lack of variation in edaphic properties great enough to make functional differences relevant for fitness. Logically, in communities where variation in edaphic properties does not change community structure or function, a stochastic relationship should be found as environmental filtering would not occur. Likewise, our observations may be skewed due to our time of measurement where specialists are using compounds that generalists metabolized at the beginning of inoculation into EcoPlates.

Discerning how soil microbial communities are structured and function is essential given their importance to biogeochemical cycling. In this study, we explored the relationship between edaphic properties and the soil microbiome. A wide range in microbial function was observed and was redundant across the site; this contrasts with differences in taxonomy and suggests functional redundancy. In future studies, setting up experiments at small spatial scales and across temporal dimensions will be necessary to better discern how stochastic processes affect community assembly.

ACKNOWLEDGMENTS

We would like to acknowledge Cole Calderon, Emily Hooser, Ana Hernandez and the Wild Basin Wilderness Preserve staff for their assistance. We thank the reviewers and editor for their time and energy. We also thank Caroline Polachek for all the hours of music listen to while working on this manuscript.

FUNDING

This work was supported by the Dr Allan W. Hook Endowed Wild Basin Creative Research Grant and the St Edward's University Biology Department.

Conflict of Interest

None declared.