Spatial analysis of a hydrocarbon waste‐remediating landfarm demonstrates influence of management practices on bacterial and fungal community structure

Summary Cultivation of dedicated soil plots called ‘landfarms' is an effective technology for bioremediation of hydrocarbon waste generated by various industrial practices. To understand the influence of soil conditions on landfarm microbial communities, analysis of bacterial and fungal community structure using next‐generation sequencing at different sections and depths was performed across a hydrocarbon‐waste landfarm in Regina, Saskatchewan, Canada. While a core set of hydrocarbon‐associated bacterial and fungal taxa are present throughout the landfarm, unique bacterial and fungal operational taxonomic units are differentially abundant at sections within the landfarm, which correlate with differences in soil physiochemical properties and management practices. Increased frequency of waste application resulted in strong positive correlations between bacterial community assemblages and elevated amounts of oil, grease and F3 – F4 hydrocarbon fractions. In areas of standing water and lower application of hydrocarbon, microbial community structure correlated with soil pH, trace nutrients and metals. Overall, diversity and structure of bacterial communities remain relatively stable across the landfarm, while in contrast, fungal community structure appears more responsive to soil oxygen conditions. Results are consistent with the hypothesis that years of bioremediation activity have shaped microbial communities; however, several management practices can be undertaken to increase efficiency of remediation, including the removal of standing water and soil tilling across the landfarm.


Introduction
Soils naturally contain diverse microorganisms capable of degrading complex hydrocarbons arising from industrial activity. Controlled application of industrial waste to topsoil has provided reliable bioremediation to reduce hydrocarbon waste in various industries (Khan et al., 2004). These treatment plots are referred to as 'landfarms', and the development of best practices for the maintenance of these systems for ex-situ treatment of wastes is attractive as a cost-saving and environmentally responsible form of waste disposal (Rhykerd et al., 1999;Besalatpour et al., 2011). Like other bioremediation technologies, treatment efficacy of landfarms depends on multiple physical, chemical and biological features, including land area, soil aeration, nutrient availability, pollutant mobility and toxicity, and microbial biodiversity (Rhykerd et al., 1999;Straube et al., 2003;Khan et al., 2004;Harmsen et al., 2006).
Effective landfarm management practices will consider and modulate these soil features to improve the activity of the bacterial and fungal communities, which are the major drivers of the biological degradation of hydrocarbons and other waste compounds applied to landfarms and other remediation technologies (Leahy et al., 2003;Straube et al., 2003;Seo et al., 2009;Militon et al., 2010). These studies focused on the effects of operational parameters such as tillage, waste dosing/application rates and moisture content on overall remediation efficacy, but did not consider the effects of management practices on microbial community structure and variability. Alternatively, several studies have examined the structure of bacterial communities that remediate specific pollutants, either using terminal restriction fragment length polymorphism (TRFLP) analysis of 16S rRNA genes present in a petroleum refinery landfarm (Katsivela et al., 2005) or denaturing gradient gel electrophoresis (DGGE) analysis of 16S rRNA genes to investigate the bacterial community in a lubricant and diesel oil-polluted landfarm microcosm . While these studies have contributed to understanding the nature of landfarm bacterial communities, no study has yet to perform extensive or high-resolution analysis of both bacterial and fungal communities throughout an operational landfarm.
Here we apply high-resolution next-generation sequencing to profile the bacterial and fungal taxa of a long-term operational landfarm in Regina, Saskatchewan, Canada, and analyse community structure and diversity in response to hydrocarbon application, soil depth and soil saturation. Such an established landfarm is expected to contain mature microbial communities selected for their ability to metabolize complex hydrocarbons, and thus presents an ideal opportunity to characterize the response and stability of microbial communities to landfarm management activities.

Landfarm soil characteristics
The EVRAZ Inc. landfarm is divided into three separate management sections, A, B and C surrounding a waste holding area (Fig. 1), with the collected hydrocarbon waste being applied in late spring (Table 1). Soil samples were collected at two depths, 'surface' (0-15 cm) and 'deep' (15-30 cm) from ten sites of landfarm sections (and once from nearby borrow-pit, ERW; Fig. 1), following tilling of the site in June 2015, with 'surface' samples collected again in November (Table 1). Multiple features of soil chemistry are routinely measured as part of landfarm management, providing critical insights into the abiotic features of the landfarm soils. Section A is characterized by features important for robust biological activity: a near neutral pH, the highest moisture and the highest concentrations of nitrogen and carbon (Fig. 2). Petroleum-related hydrocarbons BTEX plus the Canadian Council of Ministers of the Environment (CCME) hydrocarbon fractions F1 (number of carbons ranging from: C6-C10) and F2 (C10-C16) were below detection limits across all sites, except for trace amounts of F2 hydrocarbons detected in section C surface soils (Fig. 2). The long CCME carbon chain F3 (C16-C34) and F4 (C34-C50) fractions are more recalcitrant hydrocarbons and include polycyclic aromatic hydrocarbons. Both landfarm sections A and C received fresh applications of hydrocarbon-contaminated clays in April 2015, 2 months before sampling in June 2015 (Table 1), contributing to the high concentrations of fractions F3 and F4 in sections A and C. In section A, total carbon, oil and grease (which includes non-water-soluble oils and greases), and the CCME hydrocarbon fractions were most abundant in surface soils, whereas section C demonstrated the converse pattern of higher hydrocarbon levels in deep soils (Fig. 2).

Landfarm microbial diversity
To survey microbial diversity, the 16S rRNA gene (bacterial) and ITS-2 region (fungal) taxonomic markers were sequenced for all soil samples. Sequencing error rate was 0.002%, as determined by the mock community, and Goode's coverage estimates, which are the proportion of DNA sequences belonging to operational taxonomic units (OTUs) that have been observed more than once, averaged 0.956 for bacterial and 0.990 for fungal communities, indicating sufficient DNA sequencing coverage to profile microbial communities in all samples. Up to 1200 bacterial and 400 fungal OTUs were identified in each sample, with a total of twenty-three bacterial and 7 fungal phyla identified in the landfarm. A generalized linear model (GLM) demonstrates that bacterial diversity is significantly higher in surface soil within landfarm sections A (effect estimate À1.557; P < 0.00001) and B (effect estimate -1.2044; P = 0.0254) ( Table 2), with section A receiving more frequent waste application (Fig. 3). Alternatively, it is expected that the deep section B soils are the most anaerobic within the landfarm, given the soil-saturating presence of standing water, which corresponds to the statistically modelled difference between the surface and deep samples of section B (Table 2; Fig. 3). Interestingly, there was no significant difference in fungal diversity between surface or deep soils within any of the three landfarm sections (Table 2). However, an effect of standing water on section B fungal communities was observed when we examined differences in diversity between sections by combining surface and deep samples from each section. By this measure, bacterial diversity was not significantly different across the landfarm, but fungal diversity was significantly lower in section B compared to sections A (effect estimate 0.62161; P = 0.00146) and C (effect estimate À0.92973; P < 0.00001) ( Table 2; Fig. 3B).

Landfarm microbial taxonomy
Differential abundance analysis of bacterial and fungal OTUs across landfarm sections demonstrate that landfarm sites are distinct from the control ERW site (Fig. 1), which is presumed to represent the natural and ancestral state of the landfarm soils ( Fig. 4A and B). When considering the 50 most abundant bacterial OTUs, landfarm sections do not form natural clusters (Fig. 4A), revealing that despite the similarities in overall diversity described above, there is heterogeneity among the most abundant taxa between sections and at both depths within sections. Section B samples contained the most distinct OTU abundance profiles, as section B surface and deep OTUs did not cluster with each other nor closely to the other sections and control soils. Multiple classes of bacteria that metabolize hydrocarbons in the absence of oxygen were found only in the water-saturated soils of section B. For instance, Anaerolineaceae, which include anaerobic bacteria associated with n-alkane degradation (Liang et al., 2016), was abundant and common, particularly in deep soils (Fig. 4A). Although it did not rank in the few most abundant illustrated in Fig. 4A, Hydrogenophilales (Thiobacillus), a microaerophilic hydrogen-oxidizing bacterial group (Orlygsson and Kristjansson, 2014), that was also particularly abundant in section B (Table S2).
Across the landfarm, common soil residents Actinomycetales were the most abundant order of bacteria (Table S2) and soils were highly enriched in taxonomic groups identified in other hydrocarbon-contaminated soils and water (Peng et al., 2015), consistent with the hypothesis that EVRAZ landfarm soils communities have been refined by years of bioremediation activity. These include (but are not limited to) bacterial taxonomic groups of Pseudomonadales (Leahy et al., 2003;Seo et al., 2009), Acidobacter (Seo et al., 2009;Peng et al., 2015), Gammaproteobacteria and Actinobacteria (Militon et al., 2010;Peng et al., 2015), Chloroflexi, Planctomycetes and Bacteroidetes (Peng et al., 2015) ( Table S2). Specific genera previously demonstrated to be associated with hydrocarbon-impacted sites and high molecular weight polyaromatic hydrocarbons, such as bacterial OTUs identified as Anaerolineaceae, Caldilineaceae (Zhang et al., 2017) and Lutibacter (McFarlin et al., 2018) are shown to be elevated in abundance in samples from shallow and deep soils of A and B relative to section C (Fig. 4A). Additionally, hydrocarbon-associated OTUs Terrimonas (Adrion et al., 2016) and Agromyces (Zhang et al., 2010) were found abundant in section A soil, while Thiobacillus (Militon et al., 2010), and Petrimonas (Grabowski et al., 2005), were found in higher abundances in section B soil.
When fungal OTU abundance is considered, surface and deep samples cluster together within each section, while ERW forms the most divergent cluster (Fig. 4B). Ascomycota were the most abundant order of fungi (Table S3), and fungal hydrocarbon-related taxonomic  (Table S3). Specific OTUs belonging to Sordariomycetes previously demonstrated to be associated with hydrocarbon-impacted sites and high molecular weight polyaromatic hydrocarbons (Marchand et al., 2017) are increased in abundance in deep A and B section soils relative to section C (Fig. 4B).

Influence of soil parameters on community structure
Redundancy analysis (RDA) provides a powerful approach to examine associations between microbial community structure and soil characteristics. RDA of fungal OTUs showed a tighter clustering of individual samples within sections, and greater spatial distance between sections compared to bacterial OTUs ( Fig. 5A and B), in agreement with the clustering observed in Fig. 4A and B. For both bacterial and fungal OTUs in their ordination space, CCME fractions F3 and F4, Total Carbon (TC) and Total Nitrogen (TN) are all negatively correlated with trace nutrients (Ca, K, Na) and soil pH ( Fig. 5A and B). For bacteria, CCME F3 and F4, TC and TN are strongly correlated with the centroid of section A samples and the centroid of all surface samples, whereas for fungi, they are most strongly correlated with the centroid of all surface samples ( Fig. 5A and B). In both instances, these parameters are negatively correlated with the centroid for section B samples. Further, trace nutrients and soil pH are strongly correlated with the centroid for section B samples, when there is no such strong correlation for fungal OTUs of section B ( Fig. 5A and B). For both bacterial and fungal ordination species, moisture is more closely associated with section A samples, while oil and grease is negatively correlated with metals (As, Cd, Co, Fe, Mn and Mo) in the bacterial OTU ordination space, and to a lesser degree in the fungal OTU ordination space ( Fig. 5A and B).

Discussion
High-resolution sequencing of microbial taxonomic markers has facilitated spatial analysis of bacterial and fungal communities in an active hydrocarbon-remediating landfarm. Results suggest that the sustained application of hydrocarbons and management practices at the EVRAZ landfarm have selected specialized microbial communities that are highly distinct from microbial communities in a nearby soil borrow-pit (ERW) used as a control (Figs 1 and 3). Because of the regular application of industrial hydrocarbon waste to the farm soils along with the removal of plant life, we anticipated that the landfarm would be characterized by simple and highly refined microbial communities dominated by specialist species, but surprisingly the landfarm communities are composed of a greater diversity of OTUs than the neighbouring ERW pit soil (Figs 3 and 4). The microbial communities of the landfarm are comprised of taxonomic groups that have been previously associated with hydrocarbon-impacted and hydrocarbon-remediating soils. This pattern was consistent in samples collected at the beginning (June) and end (November) of the landfarm operational season, with no statistical influence of month on microbial community structure detected by the GLM (Table 2). This stability confirms the hypothesis that seasonal changes and operational activities such as tillage, fertilization and hydrocarbon application are not disruptive over time, but rather work to actively maintain the communities that have established under these management conditions. The present study provides data that supports the expectation that landfarm maintenance practices of both tillage and frequency of waste application influence the structure and profile of resident communities. The increased loading of hydrocarbon waste and fertilization at the surface of section A is what likely selects for specific hydrocarbon-related OTUs of Agromyces, Micrococcacea and Terrimonas in A relative to B and C ( Fig. 4A and B). Alternatively, Section B, (where the presence of standing water and lack of deep tillage (30 cm) favours an anaerobic environment), displays a noted enrichment of several fungal OTUs, not identified to genus level (Fig. 4B). The significant influence of water level on soil physiochemical properties and microbial community structure of non-  Table S1 are presented in graphical format for landfarm sections A, B and C, in surface and deep sampling depths collected in June. Raw and additional data are included in Table S1.  (Brockett et al., 2012;Ma et al., 2018). Thus, the selective presence of hydrocarbon waste, combined presence of standing water and lack of tillage at depths below 10 cm contribute to landfarm community structure. We observed that bacterial communities are more diverse across landfarm sites and depths compared to fungal communities ( Fig. 3; Table S2). Fungal genera are naturally highly heterogeneous and metabolically specialized (Karigar and Rao, 2011;Hewitt et al., 2016), which could explain the observation that the presence and abundance of specific fungal genera and OTUs are more greatly influenced by differences between site conditions compared to bacterial assemblages (Figs 4 and  5). RDA was used to assess the influence of these parameters on bacterial and fungal community structures, and we found that strong correlation occurs with CCME-defined F3 and F4 fractionations to all surface samples for fungal OTUs, and with all section A samples for bacterial OTUs (Fig. 5). As F3 and F4 molecules are typically more difficult to degrade, even with increased aeration at the soil surface (Seo et al., 2009), it stands to reason that the frequent application of hydrocarbon at section A contributes to increased concentration of these recalcitrant products in surface soils (Fig. 2). Similarly, less effective microbial assemblages and lower aeration may explain the accumulation of F3 and F4 molecules in the section C sub-surface soils. Thus, while complete remediation of F3 and F4 by section A microbial communities may still require extended treatment periods, RDA suggests that the differentially abundant bacterial and fungal OTUs of section A are uniquely adapted to higher concentration of recalcitrant hydrocarbons. Surprisingly, section A surface samples demonstrate increased moisture content compared with section B (Figs 2 and 5), suggesting a positive relationship between hydrocarbon waste and retention of soil moisture. This finding provides context and support for previous analysis indicating that remediation activity is optimal with increased soil moisture, in addition to aeration (Bahmani et al., 2018).
These results suggest that effective management practices would include removal of standing water from section B and spread section A soils to seed sites B and C, to maintain and potentially increase the efficiency of bioremediation. Furthermore, the more anaerobic deep soil B, which also contains specific hydrocarbon-associated OTUs, if spread throughout section B, and potentially to other sections, may contribute to increased remediation activity. Ultimately, it is evident that management practices and geochemical parameters contribute to the diversity and stability of microbial community structure, and that current conditions of the operational landfarm maintained by EVRAZ supports bacterial and fungal communities capable of contributing to effective remediation of hydrocarbon waste.

Experimental procedures
Site details EVRAZ Inc. North America (hereafter EVRAZ), is a leading manufacturer of tubular steel products located in Regina, Canada. EVRAZ has maintained a landfarm since 1994 for remediation of hydrocarbon spills, which includes lubricants, fuels, solvents, hydraulic fluids and antifreezes used during steel manufacturing. The landfarm is located in the Dark Brown Soil Zone of the Western Canadian Prairies. Soils here are described as dark brown, Vertisolic soils with very fine, uniform, calcareous lacustrine deposit parent material. Routine soil sampling of the landfarm indicates that toxic hydrocarbons from applied waste are degraded in the first few centimetres of soil; a success which has been attributed in some part to the landfarm maintenance programme that includes nitrogen fertilization, aeration through tilling, regular removal of standing water, occasional application of sanitary sewage treatment plant sludge and pH monitoring (Schutzman, 2014). OTU abundance values are centred across rows. Heatmap legend indicates result of unit variance scaling, where the standard deviation from a mean (0) is calculated across all columns and can be compared for each OTU between sample type. Both rows and columns are clustered using correlation distance and average linkage. For each fungal OTU in (B), the most refined taxonomic annotation is provided, with p_; c_; o_; f_; g_ indicating phylum, class, order, family and genus, respectively, as few OTUs are able to be identified at the genus level. There are three separate management sections of the landfarm, A, B and C, which surround a waste holding area (Fig. 1). Hydrocarbon waste is continuously collected from steel plant activities and stored over the course of fall, winter and early spring months (Table 1). Sections A and C share similar soil characteristics; however, section A has historically received more waste than section C. Section B has the lowest hydrological grade, with water accumulating in this section after snowmelt or heavy periods of summer precipitation. In 2015, application of accumulated waste to sections A and C began in April and occurred periodically until late November. After the majority of standing water was removed, in early summer and fall, soils were tilled, followed by fertilization with urea in November (Table 1).

Sample collection
Samples were collected with an AMs Frozen Soil Powered Auger Kit at two depths, 'surface' (0-15 cm) and 'deep' (15-30 cm), at ten sites sections (A, B, and C) of the landfarm, following tilling of the site in June 2015, with 'surface' samples (0-15 cm) collected again in November (Table 1). For characterization of nonimpacted soils, an additional sample was taken by gloved-hand using gloves from the top 5 cm of soil in an area~500 m outside the landfarm for characterization of non-impacted soils (ERW borrow-pit; Fig. 1). From bagged soil samples for each depth, approximately 50 g was subsampled by gloved-hand and processed for bacterial and fungal community analysis. Soils collected from each of the ten individual sites per landfarm section were analysed via sequencing of taxonomic gene markers as independent samples and not composited. The remainder of the June samples were composited into landfarm sections (A, B and C) at each depth and sent to Saskatchewan Research Council Environmental Analytical Laboratories (Saskatoon, SK) for routine analysis of pH, trace micronutrients (Ca, K, Na), metals (As, Cd, Co, Fe, Mn, Mo), total organic carbon, total nitrogen, phosphorus, oil and grease, and hydrocarbons: benzene, toluene, ethylbenzene, and xylene (BTEX) and CCME hydrocarbon fractions F1 -F4 (Fig. 2 and Table S1; CCME, 2008).

Microbial community profiling
Environmental DNA was isolated from 0.25 g of homogenized soil samples using MoBio Powersoil DNA Isolation (Carlsbad, CA) according to manufacturer's instructions. Bacterial and fungal sequencing libraries were prepared according to Kozich et al. (2013), using primers designed for the v4 hypervariable region of the bacterial 16s rRNA gene, and primers for the fungal internal transcribed sequence (ITS-2) region (Schoch et al., 2012;Gweon et al., 2015). Sequencing of a bacterial mock community (BEI Resources; Manassas, VA, USA) facilitated calculation of the sequencing error rate (Kozich et al., 2013). Sequencing was performed on the Illumina MiSeq platform for both bacterial libraries (V2; 250 bp paired-end reads) and fungal libraries (V3; 300 bp paired-end reads).
Prior to all subsequent analysis, data were rarefied to the level of 5,307 and 11,077 sequences for bacterial and fungal samples respectively. Inverse Simpson diversity was calculated for all samples and then analysed using a GLM to determine the effect of section, depth and month of sampling as factor covariates. An interaction between landfarm section and soil depth was included in the GLM. Sample diversity is a positive real-valued response variable with strong mean-variance relationship. To account for this, we assumed the response was conditionally distributed following a Gamma distribution with support on the positive real values. Generalized linear hypothesis tests of differences between selected treatment effects were performed using contrast coding and P-values adjusted for multiple comparisons via the multcomp package (Hothorn et al., 2008) for R (v 3.5.1; R Core Team, 2018). Abundance of bacterial and fungal taxa from all collected samples was visualized via ClustVis (Metsalu and Vilo, 2015).

Redundancy analysis for soil chemistry correlation
To assess relationships among geochemical parameters and community structure, RDA was performed on select parameters of the geochemical data set ( Fig. 2 and Table S1), using the vegan package (v.2.41; Oksanen et al., 2016) for R. Bacterial and fungal OTU data were screened to remove taxa present in fewer than five samples in the data set and subsequently Hellinger transformed (Legendre and Gallagher, 2001). The month of sampling was used as a partial covariate to remove the effect of sample collection date from the data prior to analysis. The direction and magnitude of greatest change in RDA ordination space for each geochemical variable were calculated using a least squares regression. Each geochemical variable, in turn, was used as the response variable, and the coordinates of the samples on RDA1 and RDA2 were used as covariates. The two resulting regression coefficients provide vector coordinates in the RDA space, which represent the direction of maximal change. The R 2 statistic provides indication of the strength of the correlation between the configuration of samples in ordination space and each geochemical variable (assuming a linear relationship). Correlations were assessed statistically using a randomization test with 999 permutations.

Supporting information
Additional supporting information may be found online in the Supporting Information section at the end of the article.