Rothamsted Repository Download

: We use a unique set of terrestrial experiments to demonstrate how soil management practises result in emergence of distinct associations between physical structure and biological functions. These associations have a significant effect on the flux, resilience and efficiency of nutrient delivery to plants (including water). Physical structure determining the air-water balance in soil as well as transport rates is influenced by nutrient and physical interventions. Contrasting emergent soil structures exert selective pressures upon the microbiome metagenome. These selective pressures are associated with the quality of organic carbon inputs, the prevalence of anaerobic microsites and delivery of nutrients to microorganisms attached to soil surfaces. This variety results in distinctive gene assemblages characterising each state. The nature of the interactions provide evidence that soil behaves as an extended composite phenotype of the resident microbiome, responsive to the input and turnover of plant-derived organic carbon. We provide new evidence supporting the theory that soil-microbe systems are self-organising states with organic carbon acting as a critical determining parameter. This perspective leads us to propose carbon flux, rather than soil organic carbon content as the critical factor in soil systems, process-form relationship.


Introduction
Soil -the basis of terrestrial life on Earth -continues to defy our comprehensive understanding despite the evident catastrophic consequences of mismanagement, such as 40 the North American "Dust Bowl" of the 1930's which was exacerbated by poor stewardship of agricultural soils [1]. Faced with the multiplicity of processes which constitute soil, scientific reductionism has led to studies which have advanced our knowledge of soil's biological, chemical or physical components predominantly in isolation. However, soilin common with many biological phenomena -is more appropriately considered a 45 hierarchical assemblage of interacting processes, stabilized and actively maintained at different timescales [2]: soil is processual and not comprehensible based on singlediscipline experimentation. Tisdall and Oades' pioneering conceptual model [3] linking microbial activity to soil structural development advanced the importance of interaction between biotic and abiotic phenomena in the process of generating soil structural 50 complexity (topology and connectivity).
Soil organic matter (SOM) is the fundamental causative agent generating structural complexity, as it acts to bind mineral particles and colloids together. Plant and animal residues are processed by microbes before joining the SOM pool [4,5]: this step is an important facet of both the Tisdall and Oades model, and its subsequent extension [6].

55
SOM may take the form of microbial polysaccharidic and proteinaceous exudates as well as cell debris and is chemically structurally diverse [4]; in effect, SOM is a continuum of progressively more extensively oxidized compounds [7]. Much of this SOM is associated with pores of 30 -100 μm diameter [8], scales comparable to the 12 -13 μm distances observed in soil between microbial cells [9]. As a result, the effect of microbial processes 60 -metabolism, extracellular degradation of compounds, polymer secretion and cell lysison soil structure is particularly evident at the scales < 50 μm responsible for regulating convective and diffusive flow rates, as well as the balance of air and water at any given matric potential [6]. These hierarchical processes exhibit characteristic properties of selforganizing and emergent systems [10,11]. 65 Such experiential and theoretical approaches are formulating a new understanding of how microbial activity controls soil structure -in effect, how soil should be viewed as an expression of biological process. They also provide evidence supporting a view of soil as a product of genes, manifest through the combined effects of multiple organism phenotypes: in essence, an extended composite phenotype (Phillips [12], after 70 Dawkins [13]). The identifying features of this phenomenon are a strong influence of at least one organism upon the form or structure of a soil environment -termed a processform relationship; demonstrable synchrony between the activity of influencing organisms and form development; selective pressure arising from form development acting, in Dawkins' strict sense upon alleles [13,14] and in Phillips' broader concept upon soil 75 organisms [12]; which results in positive feedback where selective pressure favours alleles (or organisms) associated with the process-form state, manifest as the influence of microbial turnover of SOM upon soil structural development, discussed above.
There is compelling evidence implicating plant-derived organic carbon inputs in the soil extended composite phenotype [15,16]. However, complete description of such a 80 phenotype requires, in turn, a well-developed understanding of the consequences of evolving soil structure for the genetic manifestation of on-going microbial processes -such feedback is necessary for emergence of organisation, observable at the whole-system level in complex biological, chemical and physical systems. Currently, few studies present comprehensive description of the influence of soil structure upon microbial processes, and 85 those that do, typically address only the association of metabolically defined bacterial groups with soil aggregate or particle size, rather than soil structure per se (see Lensi et al., [17] and Chotte et al., [18]). The principal influence of soil structural complexity is predicted to be on diffusion processes dictating the microenvironments surrounding surface-associated cells [19]. Observation of anaerobic regions of soil aggregates 90 associated with denitrification processes [20], and the influence of anaerobic microsites in ostensibly oxygen-rich soils upon microbial respiration and carbon compound oxidation rates [21] provide indirect evidence for such metabolic constraints arising from soil structure. However, this view of soil as an extended composite phenotype requires two specific conditions to be met. The first we term the Process-Form Condition, where the patterns and whole metabolic pathways) such that alleles that correspond with specific processes are preferentially selected for -extending beyond short-term quantitative changes in specific gene expression profiles.
In this paper, we integrate biological and physical data relating to dynamics of the soil system with mathematical modelling to explore these conditions. This approach is 105 used to interpret results from a unique long-term field-experiment within the context of the proposed view of soil as an extended composite phenotype: linking organic carbon inputs to soil with emergence of key soil structural properties; and describing the genelevel microbiome responses to contrasting emergent soil structural complexity arising from long-term carbon input regimes. The experiment uses the Highfield Ley-Arable 110 Experiment at Rothamsted Research, Harpenden, U.K.

Results
Process-form relationships in soil are expressed through fine-scale connected porosity. We first investigated the influence of added organic carbon (C org ) on the development of soil structural complexity, testing the hypothesis that greater inputs of 115 C org to soil are associated with development of improved soil structure; assessed as the degree of connectedness between pores (connected porosity). The bare fallowed soil used as a starting point for these experiments experienced a demonstrable decline in C org [22] and microbial abundance [23] over forty-eight years of continuous management.
Estimates of C org are approximately 3 g-C kg −1 [24] and the soil has a significantly reduced 120 total porosity compared to mixed grass sward soils [25]. To assess the influence of newly imposed managements, we followed structure development for a decade (2008 -2018)  and grassland soils were also modelled in the same manner but using their respective starting dates. The extremes of the cumulative C org input-connected porosity relationship ( Fig. 2) are derived from consistently managed soils; bare fallow being associated with the lowest net C org input and connected porosity, and arable and grassland soils being 150 associated with the second highest and highest C org input and connected porosity, respectively. Modelled C org inputs and measured connected porosity for soils converted to both arable and grassland for each of the ten years between 2008 and 2018 is distributed between the consistently managed bare fallow and arable soils data. We assumed that C org in soils managed as grassland since 1838 represented the maximum which could be 155 stored; this and the Akaike information criterion was used to guide selection of a sigmoidal function. There was a clear non-linear relationship between C org inputs to soil and connected porosity, with all converted and continuously managed soils following the same trend (Fig. 2). This establishes that process-form relationships can be explained in terms associated with biotic C org inputs and turnover in soil.

160
Contrasting long-term soil management results in different process-form states. Although the converted soils provided compelling evidence for structural development contemporaneous with the establishment and development of plant populations (albeit that arable soils are subject to external processes such as tillage and fertilization), the relatively short time span did not allow comparison of maximal 165 differences in structural development, or evaluation of the potential for any resulting selective pressures to influence the representation of specific genes within the soils. Soils which had been under continuous management, at the time of sampling, as bare fallow for fifty-two years, arable for sixty-two years and mixed grass swards for over two hundred years presented an opportunity to test the hypothesis that established process-form 170 relationships result in selection of organisms or genes, the fitness of which is suited to each particular soil biotope.
We have already determined that the continuously managed grassland and arable soils have significantly greater total porosity, a wider range of pore sizes and greater pore connectivity than continuously managed bare fallow soil [25]. Here we extend these the maximum pore throat size of connected pores controlling hydraulic conductivity [27].
Topology-related parameters derived from X-ray computed tomography of aggregates (Table I)    the soil pore space. Fig. 3 shows the combined effects of soil C org and connected porosity upon the predicted hydraulic conductivity of soils. Combined direct measurements and modelling indicate a power law relationship between connected porosity and conductivity and that C org is associated with these changes. Regions of this relationship correspond to the process-form states of continuously managed bare fallow, arable and grassland soils.

210
The fraction of anoxic volume characterising each process-form state was estimated using a multi-phase lattice-Boltzmann approach [28,29], described in detail in the Materials and Methods section. The results (Fig. 4) indicate that the predicted anoxic fraction is significantly lower in grassland soil, compared with arable and bare fallowed soils, the latter is predicted to have the highest fraction of anoxic volume at all matric potentials 215 (moisture contents).
Microorganisms in each land management-associated process-form state are likely to experience markedly different hydraulic environments, particularly in degraded bare fallow soils where reduced delivery of dissolved nutrients and O 2 is predicted compared to grassland soils. This is a direct result of constraints placed upon diffusive flow by reduced 220 connected porosity and d crit resulting from different biotic C org inputs and turnover. These constraints are likely to exert significant selective pressures in soil microbiomes which should be reflected in changes in the assemblages of organisms or alleles in the different process-form states. To test this hypothesis, we generated shotgun metagenome sequence datasets from nucleic acids extracted directly from the different soils. These were 225 analysed to determine whether any observed differences in phylogenetic community assemblages or in gene abundance were directly attributable to the differences in porosity or d crit described above. There was again a significant treatment effect upon S Chao1 (F 2,6 = 11.8, p = 0.008) and pairwise comparison indicated grassland was significantly more rich in fungal OTUs than either arable or bare fallowed soils (smallest difference, Q = 5.68, p = 0.017), but there  Figure 2) using KR distance. The first two principal coordinates separated treatments clearly, the ordination accounting for 89% of total variation across the two axes. Distance-based linear modelling (distLM) was used to describe the relationship between the 16S rRNA gene-conditional phylogenetic community structure (using KR distance) and edaphic variables shown in Table II. All Although there are clear qualitative and phylogenetic differences between prokaryote assemblages associated with each soil, distLM suggests that these differences are adequately described by chemical edaphic parameters. They are therefore unlikely to 295 be due to selective pressure arising from the respective process-form states directly.
Instead, assemblage differences are likely to reflect organism traits: for example, Gemmatimonadetes are common in soil and show adaptation to low soil moisture [33], so identification of Gemmatimonas as characteristic of bare fallow soil is likely to reflect the fact that direct isolation experienced by these soils renders them much drier than the 300 other soils; nitrogenous fertilization of arable soils is reflected by the organisms identified as characteristic of these soils to be either nitrite-oxidisers or denitrifiers; and while identification of Rhizobiaceae as characteristic of the mixed swards of grassland soils suggests association with legumes -and therefore possibly responsive to selection pressure exerted by the plant population, Bradyrhizobium spp. in these soils lack genes 305 and gene clusters for symbiosis and nitrogen fixation [34].
Process-form states in soil exert selective pressure at the level of alleles. A total of 1,197 KEGG orthologs were identified as having significantly different abundance between the soils (presented in detail in the Supplementary Fig. 3 -7). We adopted a similar approach to analysing the effects of soil management upon microbiome genetic shown to be sensitive to biotic C org inputs and turnover in soil ( Fig. 2 and 3). We have also demonstrated that they exert a dominant influence upon hydrodynamic conductance of 335 the pore network and the potential for anaerobic sites across a range of soil matric potential ( Fig. 3 and 4). However, the analysis cannot demonstrate preferential selection of genes dependent upon their fitness within each process-form state. To test whether the differences in gene abundance could be due to selection pressures arising from different plant inputs and emergent soil structural properties, we characterised the genes shown Other responses, such as the increase in gene abundance for chemotaxis and protein 375 secretion, may also be responses to reduced diffusion of soluble nutrients, and hence a requirement to search out nutrients, or avoidance of anaerobic niches within the soil.
Microbial community structure is often considered as a balance of cooperative behaviours between individuals, mediated by "public goods" or soluble nutrients arising from leaky processes (nutrients which are lost through the outer membrane or released by cell lysis)

380
or the activity of exoenzymes [35,36]. Producers of public goods support populations of "cheaters" which exploit goods without contributing to them. In well-mixed systems, cheaters maintain a competitive advantage over producers, but this advantage is lost in structured environments where diffusive constraints are manifest [37]. In this context, the increase of T2SS and arabinanase and chitinase exoenzyme genes in arable and bare 385 fallowed soils may be a response to both qualitative changes to organic inputs, and reduced delivery of soluble nutrients by advective flow and diffusion to cheaters, and thus an increase in abundance of producer organisms. Additionally, the reduced diffusive processes predicted for arable and particularly bare fallowed soil may result in an increased efficiency of exoenzymes since reduced diffusion allows for a greater 390 accumulation of product near producer organisms [38]. Thus, production of exoenzymes, and cell motility as a searching or avoidance behaviour provide adaptations in response to spatially constrained circumstances arising from reduced pore connectivity as a result of reduced C org inputs in arable and bare fallowed soils.

395
We have presented data consistent with the conditions that should be met if soil is an reflecting the emergence of coherence across a wide range of scales (often referred to as fractal scaling). In Fig. 3 we observe such behaviour between connected porosity and hydraulic conductivity. The state of the soil system changes from one with a disconnected pore space to one with a connected pore space where C org (energy) flux is a critical 415 parameter. There is a power law relation between conductivity and porosity, consistent with the emergence of large-scale spatial coherence in soil structure at a critical value of C org flux. In this sense, soil displays many of the properties of self-organising systems [39].
Results presented here provide further evidence for a causal feedback between allelic abundance, process and form. We have previously posited a mechanism for this in soil 420 and shown how soils with and without plants are capable of spontaneously generating emergent structures at important scales [6] compared with sterile soils, which do not. This interpretation predicts that soils which are more self-organising will be more metabolically active in any given situation than a soil where the interaction between biological process and form is weak or non-existent. We see that, after a minimum of fifty-two years, each soil in our study is a different expression of its multiple biotic components; a phenomenon termed an extended composite phenotype [12].
With plants present, such as land managed as long-term mixed grass sward, the extended composite phenotype has an increased capacity to store water and soluble nutrients, a property which may confer a degree of resilience to the soil-plant-microbe 430 system during periods of low rainfall or nutritional inputs. Independent analysis of these same soils has demonstrated greater water storage capacity in the grassland soils [40].
In addition, the more extensive and more connected pore network selects for assimilatory, The finding that soil under grassland management has significantly higher 440 capacity, efficiency and resilience compared with arable or bare fallowed management is associated with greater C org inputs and turnover. Furthermore, the rate of recovery of degraded soil is also linked to stocks and flows of C org (Fig. 1). Our experiments cannot distinguish between C org flux or storage as the dominant mechanism supporting improved soil function. However, interpreting results in terms of soil remodelling through self-445 organizing processes, we predict that the biophysical state of soil and rate of change of that state will both be related to cumulative metabolic activity. Our data are consistent with recovery rate being limited by cumulative soil metabolism: soil C org content acts as a diagnostic for this. This raises the important questions of what limits soil metabolism and incorporation of C org in soil [41], and how it can be manipulated in each context to 450 maximise the rate of soil recovery. We know both anaerobic niches and physical dislocation of microbes from resources result from low pore connectivity, and both significantly limit microbial metabolism. We also know soil recovery is associated with more voluminous and better-connected pore space and significantly lower levels of anaerobic respiration. We speculate that the rate-limiting factor in recovery of degraded 455 soil is the process of microbially-mediated micro-structure remodelling, and that this is soil texture dependent [25]. Sandy-textured soil would be less able to recover compared to soils with higher fractions of silt and clay, where remodelling fine-scale structure is inherently more feasible due to a greater proportion of "raw materials" to enable such fine-scale architecture to be manifest. It is also likely to be dependent on the quality and 460 quantity of organic inputs to soil, especially in relation to the latent energy contained in them. This is apparent in our data, though we are not able to distinguish between the relative importance of each.
Tillage is known to contribute to decreases in soil C org , and the most effective recovery rate and highest metabolizing end-state in our data was achieved with 465 management under grassland without tillage. Tillage has the effect of significantly changing the distribution of microenvironments in soil through increased aeration and exposure of previously physically protected prey organisms and soil C org . This results in the immediate release of physical and chemical constraints on metabolism and therefore to loss of soil C org . More importantly, rearrangement of microenvironmentsi.e. within 470 and between soil macro-and micro-aggregates -will have the effect of "re-setting" microbial remodelling of soil microarchitecture, slowing down establishment of connected pore space and longer-term cumulative metabolism.
This new interpretation of the role of nutritional and physical management of soil is a step towards a more general theory of soil. Such a theory is needed as a framework 475 upon which to synthesize data and knowledge on biological, chemical and physical properties of soil that are typically studied in isolation. Theory leading to quantitative prediction is also essential in seeking synergistic interventions that recognise the interplay between capacity, efficiency and resilience of soil, and to avoid the unintended consequences of land management that are directing us towards systemic collapse of 480 productive land and an amenable climate.  In addition, we also sampled plots which had been managed consistently as bare fallow for fifty-two years, arable for sixty-two years or mixed grass swards since 1838. Physical and biological data has already been reported for these consistently managed soils (Table I).

500
Over these periods, the bare fallowed soils have become depleted in more labile organic carbon and enriched in persistent organic carbon [42] and soil organic carbon has been reduced to a greater extent than in arable soil. There has also been an observable progressive shift, from grassland to arable and bare fallowed soil, in the distribution of organic carbon between different pools in the three soil managements, particularly a 505 relative decline in discrete organic particles independent of stable soil aggregates, and a corresponding increase in the proportion of organic particles encapsulated in stable aggregates [23]. Confirmation of this apparent shift in soil structure has been provided by high-resolution X-ray Computed Tomography [25].  [27]. χ(d) is a well-defined characteristic related to pore space topology and shown to be critical to hydraulic properties [43].  Supplementary Fig. 9, the temporal change in solute concentration inside any pore voxel can be calculated using the finite volume approach, as follows:

X-ray Computed Tomography and Image Analysis
where c is concentration, q is diffusive flux, D is molecular diffusion of the solute in liquid ( ) where D eff is the effective diffusion coefficient, N is the total number of pore voxels in the simulated images, ( ) z i q x is the vertical diffusive flux in pore voxel centred at location x i , 560 L z is the height of the image as shown in Supplementary Fig. 9. To address the impact of change in pore geometry due to management on the ability of the aggregate to diffuse solute, in result analysis we normalized the effective diffusion coefficient D eff of all solutes by their associated molecular diffusion coefficient in non-constrained water, D.

Modelling of Oxygen Diffusion and Anoxia -
The impact of soil structure on O 2 565 diffusion and its subsequent consumption by microbes under various saturations was studied using pore-scale simulations. We first calculated the spatial distribution and connectedness of different pores and then determined water distributions in pores under different matric potentials (ψ m ). We assumed the soil was initially saturated and then applied a negative pressure p at the bottom to drain water. We assumed the soil was 570 essentially hydrophilic in that only pores whose associated capillary pressure p c , calculated by with σ being water-air surface tension, is less than p and that / c p r = σ they form clusters which stretch from the top to the bottom of the structure can be drained. Supplementary Fig. 10A shows an example illustrating water distribution in the structure calculated using the method described above when the saturation is 55%.

575
Once the water distribution was determined for a given ψ m , we treated the water-air interfaces inside the structure as a boundary at which gaseous O 2 dissolves and then moves toward the solid-water interface to be reduced by microbial reactions. The partial pressure of gaseous O 2 in the simulated structure was assumed to be constant. Movement of dissolved O 2 in the liquid water was simulated using the following diffusion-reaction The above equation was solved by a finite volume method with each water-filled voxel being the element used to calculate the mass balance. In all simulations, water was assumed be initially free of O 2 and we simulated the system to steady state. As the 600 development of anaerobic areas was a balance between the ability of soil to diffuse dissolved O 2 and the microbial consumption rate, to elucidate that the relative anaerobicity of soils under the same ψ m is the consequence of their structures and does not change with microbial reactive rate, we simulated two scenarios: a fast microbial decomposition (kʹ= 1x10 -2 ) and a slow microbial decomposition (kʹ= 1x10 -4 ). For each 605 scenario, once the system was deemed to have reached a steady state, we sampled sites where concentration of dimension-less dissolved O 2 was less than 20% assuming them be at anaerobic condition [46]. Supplementary Fig. 10B shows an illustrative example of the location of anaerobic areas simulated by the above method in which soil particles were made transparent. We repeated the procedure to achieve different water distributions Modelling of organic carbon dynamics in soil -We used RothC-26.3 [47] to model the turnover of soil C org in the experimental soils, accounting for the effects of soil type, 615 plant cover and historical temperature and moisture content on organic carbon turnover processes. We used the same inputs of C org to the soil as those used by Johnston et al. [48].
To obtain the starting soil C org of 63.6 Mg-C ha -1 , input to the soil from plant debris, roots, and root exudates was 2.7 Mg-C ha -1 , with inert organic matter (IOM) being 3.0 Mg-C ha -1 .
The incoming C org from plant residues were assumed to have decomposable plant material paired-end reads. The generated sequences were limited to a minimum quality score of remove substandard sequences, the average metagenome size for each soil was 4.96x10 8 reads for grassland, 2.86x10 8 for arable and 2.88x10 8 for bare fallow soils. Since differences in library sizes were less than 10-fold, we did not employ rarefaction before 645 comparing the datasets [50].

765
(BBS/E/C/000J0300); Author contributions: All authors played a part in conceptualizing the study and methodology development; AB-L generated and analyzed the X-ray computed tomography of soil pore space under supervision of KR and SJM; IC manages the Highfield Conversion experiment and also performed nucleic acid extractions, library preparation for sequencing and preliminary metagenomic analysis;

770
ALN performed analyses of metagenomic data and prepared the original draft manuscript and was responsible for data presentation; XZ was responsible for programming and software development associated with modeling of hydraulic conductivity and soil anoxic volume, implementation of computer code and supporting algorithms; KC was responsible for implementation of the RothC-26.3 computer code and supporting algorithms 775 associated with modeling organic carbon in the soils; all authors participated in the development of the final version of the manuscript; Competing interests: the authors declare no competing interests; Data and materials availability: data, code, and materials described in this research are available together with extensive chemical, climate and treatment data and history on the e-RA database, maintained by Rothamsted
970 Figure 1. Grassland soils generate connected pore space more rapidly than arable soils. Degraded soil (managed as bare fallow since 1959) developed greater connected micro-porosity following conversion in 2007 to grassland than bare fallow soil converted to arable. The mean and standard error of the mean of connected porosity measured in soil aggregates collected from soil managed continuously as bare fallow 975 (brown), soil converted to arable management (dark yellow) and soil converted to grassland (green) over the ten years following conversion are shown. The dotted line marks the mean connected porosity of continuously managed bare fallow soil over the entire ten-year period.

Figure 2. Soil process-form relationships reflect biotic organic carbon inputs
980 and turnover. The connected pore space in degraded soils converted after 48 years of bare fallow management to either arable or grassland increases in association with the net input of organic carbon (C org ). Soils managed continuously as either arable (67 years) or grassland (>200 years) which have each accumulated over 100 Mg ha -1 of C org over their history follow this trend. The relationship is described by an asymptotic function; the 985 resulting fit (solid line) is shown, together with the upper and lower 95% confidence intervals of the fit (dotted lines). R 2 = 0.85. Figure 3. Contrasting long-term soil management results in quantitatively different process-form states. Soils are described by a combination of the connectivity of pore space, established from X-ray CT (connected porosity) and modelled hydraulic conductivity -a measure of capacity, representing the maximum potential movement of resources through pore networks to organisms. Grassland soils (green data points) are characterized as having high pore connectivity and hydraulic conductivity and are associated with the greatest stocks of C org . In contrast, degraded bare fallow soils (brown data points) are associated with extremely limited connected porosity and hydraulic 995 conductivity and the lowest stocks of C org . Arable soil (dark yellow) is intermediate between these two extremes. Data point size is proportional to C org (Mg ha -1 ) in each soil, the extremes of which are shown in the key. Low-C org , low-connected porosity soil contains much larger volumes of anoxic microsites than high-C org , high-1000 connected porosity soil. Across a range of matric potential (ψ m ), the predicted volume of anoxic sites is consistently larger in degraded bare fallowed soil than arable or grassland. At field capacity (θ fc ), approximately 30% of degraded soil is anoxic, falling to 5% in grassland soil. At 21 kPa degraded soil is completely anoxic while the volume remains between 4-5% in grassland soil. In arable soil 10% of the soil volume is predicted to be 1005 anoxic at θ fc -double that in grassland. Figure 5. Grassland, arable and bare fallowed soils microbial community βdiversity. Neighbour-Net networks of prokaryotic and fungal community profiles from the three soil managements based on weighted UniFrac distance. Prokaryotic community assemblages were significantly phylogenetically different between all three 1010 managements; for fungi there was no difference between arable and bare fallow soil assemblages, which were both significantly different from grassland assemblages. Figure 6. Taxonomy-based community responses to land management. A -Predictive modelling using a supervised Random Forest algorithm identified the fifteen OTUs that were most discriminatory between the different soils, based upon the mean