Australian dryland soils are acidic and nutrient-depleted, and have unique microbial communities compared with other drylands

Abstract

Aim

To compare Australian dryland soils with dryland soils globally

Location

Australian and global drylands

Methods

We used data from standardized surveys of soil properties (C:N:P content and stoichiometry, and pH), and microbes (diversity, composition and correlation networks) from Australian and global drylands, which occupy three-quarters of the Australian land mass and are the largest biome on Earth.

Results

We found that Australian dryland soils were different, exhibiting characteristics of ancient weathered soils. They had lower pH, total and available P, and total N, and greater C:N and C:P ratios than global dryland soils. Australian soils had distinctive microbial community composition and diversity, with more Proteobacteria and fewer Basidiomycota than global dryland soils, and promoted the abundance of specific microbial phylotypes including pathogens, mycorrhizae and saprobes.

Main conclusions

Australian dryland soils are clearly different from dryland soils elsewhere. These differences need to be considered when managing dryland soils to avoid unreasonable expectations about plant productivity and carbon stocks, or when predicting likely changes in ecosystem processes resulting from global environmental change.

1. Introduction

Comparisons of the biogeography of the Australian continent with the rest of the world have fascinated scientists for decades (Orians & Milewski, 2007). The literature is replete with statements referring to Australia as uniquely different in the world, being the largest island continent, and characterised by flat, barren and heavily eroded landscapes that are largely unproductive for agriculture (Andrewarthy & Birch, 1984; Orians & Milewski, 2007). Australia also supports a diverse range of unusual plants and animals such as egg-laying ‘marsupials’, and has had a long history of aboriginal occupation, but a relatively short history of European settlement coupled with European farming practices (Gammage, 2011). Unlike other continents, more than 90% of Australia’s land mass still supports native vegetation, and less than 6% is arable. Australian soils are commonly described as being nutrient-poor or infertile, and unsuitable for farming (Northcote & Skene, 1972; Lindsay, 1985; Stafford Smith & Morton, 1990; Taylor, 1994). Strikingly, despite this widely held view, and unlike what we know about the uniqueness of Australia’s plants and animals, there is a general lack of empirical evidence to support statements about the low fertility of Australia’s soils (Lindsay 1985), particularly when compared to other regions of the globe.

Drylands, ecosystems characterized by a scarcity of water, are particularly relevant when comparing Australian soils from those of the rest of world. These ecosystems occupy almost 75% of Australia’s land mass (Dunkerley, 2010) and are globally highly important, occupying about 45% of Earth’s global land area and supporting about 40% of its human population (Prăvălie,2016). At a continental scale, Australian drylands exhibit three unique characteristics compared with other drylands. First, the Australian continent is situated in the mid to lower latitudes. Therefore, unlike many other terrestrial ecosystems from the Northern Hemisphere, its soils were not strongly influenced by the last glaciation, which occurred about ~21 k ybp. Because of the lack of glacial disturbance on Australian soils, they are considered, on average, to be extremely ancient (i.e. many millions of years). For example, many of Australia’s soils had their origins in the late Cainozoic period (10-20 M ybp), where they experienced long periods of relative stability (Hubble, Isbell & Northcode, 1983). Consequently, soils from Australia are likely to have characteristics consistent with strongly weathered soils with a long history of development, including low soil phosphorus (P), low pH or low concentrations of soil carbon (C), which all contribute to the uniqueness of its soils (Wardle, Walker & Bardgett, 2014; Vitousek, Porder, Houlton & Chadwick, 2010; Laliberté et al., 2013). Second, unlike other dryland ecosystems worldwide, forests and dense woodlands dominated by Eucalyptus spp. typically occur in environments that elsewhere, would normally be occupied by drought-tolerant shrubs or grasses (Dunkerley, 2010; but see Bastin et al., 2017). These densely wooded or forested systems are often characterized by a higher amount of organic matter and lower levels of soil P and pH due to weathering related to plant productivity, contributing to the unique signature of Australian soils.

Australia has only recently been occupied by Europeans, and traditional cultures did not cultivate the land (Hubble et al., 1983). Therefore, there are almost no anthropogenic soils, i.e. completely human-produced soils resulting from direct human impact. The lack of soil rejuventation from human disturbance may also contribute to the maintainance of soils showing characteristics of strongly weathered profiles. All of these characteristics distinguish Australia from other continents where drylands form a large proportion of the land mass, such as Africa and North America. It is known that Australian soils are low in phosphorus, and soil pH levels could change in response to small changes in the water balance (Slessarev et al., 2016). Further, given the global relationship between pH and soil microbial communities (Fierer & Jackson, 2006; Lauber, Hamady, Knight & Fierer, 2009; Maestre et al., 2015), we might expect Australian soils to be characterised by a different microbial community. However, there have been no continental-wide assessments evaluating whether Australian soils and their microbial communities might differ or resemble from those from other drylands.

Here, we compared soil properties (total C and pH), soil nutrient availability and microbial communities of Australian soils with those across the globe. We pose the following question: To what extent do Australian dryland soils exhibit characteristics similar to other drylands globally? To address this question, we gathered information from five independent datasets that contained a total of 612 dryland locations and information on nutrient availability, soil C, pH and/or microbial communities globally and in Australia. We report on these soil properties because they have been used for over half a century as classic indicators of soil weathering (Walker & Syers, 1976; McGill & Cole, 1981). Similarly, soil microbial communities have recently been suggested to strongly change in response to ecosystem development (Noll & Wellinger, 2008; Alfaro, Manzano, Marquet & Gaxiola, 2017), giving us further insights into the uniqueness of Australian soils. Given the nature of the Australian landscape described above, we hypothesized that, compared with global drylands elsewhere, Australian soils should be characterised by ancient, deeply weathered soils, i.e. that are acidic and nutrient depleted. These characteristics might have also led to a very specific microbial community assembly in the continent.

Identifying whether Australian soils are really unique and and whether their microbial communities are distinct or merely a subset of those found in drylands globally is important for several reasons. First, it would fill an important gap in our knowledge that could also advance our understanding of observed biogeographical patterns of other organisms (e.g. plants). Second, we would be able to manage these soils more effectively if we have a better understanding of their inherent fertility compared with similar ecosystems worldwide that have formed under the influence of different geological and historical influences. Farming practices in Australia are still largely based on ideas imported from Europe where soils, landscapes and the response to land management practices are markedly different. Indeed, Australian ecosystems are often managed on the basis that they are assumed to be similar to other global drylands (Pickard, 1994). Understanding what makes Australian soils unique is also important to anticipate how ongoing environmental changes can affect their capacity to maintain multiple essential ecosystem functions and services (e.g. nutrient cycling, carbon storage and food and fibre production).

2. Methods

2.1. Study sites

We gathered information from five independent, large scale datasets containing information on nutrient availability, soil C, pH and/or microbial communities, which were all measured using the same protocols. These datasets included: (i) 236 sites from the BIOCOM project (Maestre et al., 2012), which included 18 Australian sites (Global Network Study, hereafter); (ii) 22 sites from a regional study across eastern Australia (Delgado-Baquerizo et al., 2015; East Australia Study, hereafter); (iii) 54 sites from a regional study of grazing impacts on eastern Australian soils (Eldridge, Delgado-Baquerizo, Travers, Val & Oliver, 2016; NSW Grazing Study hereafter); (iv) 109 samples from uncultivated dryland sites from the Australian BASE project (Bissett et al., 2016; BASE Project hereafter) available online from Delgado-Baquerizo et al., (2016) and (v) 191 soils collected as part of a regional study of biocrusts across eastern Australia (Eldridge 1999; Biocrust Study, hereafter; Supporting Information Table S1). Drylands are by definition located in areas with aridity index < 0.65, but we also included in our analyses dry subhumid areas because these areas also severely water limited, and are likely to become increasingly arid under current climate change prediction scenarios (Huang, Yu, Guan, Wang & Guo, 2016). Our study included data from soil samples collected from 612 dryland (Aridity Index 0.06 to 0.92; UNEP 1992) sites located in 19 countries (Argentina, Australia, Botswana, Brazil, Burkina Faso, Chile, China, Ecuador, Ghana, Iran, Israel, Kenya, Mexico, Morocco, Peru, Spain, Tunisia, USA and Venezuela) from six continents (Fig. 1; see Appendix S1 in Supporting Information). Moreover, this dataset included locations in temperate, arid, continental, polar and tropical climates, as defined by the Köppen climate classification (Peel et al., 2017). All soil samples were collected within the top 10 cm following standardized protocols. Data availability and references to the original protocols for each dataset are available in Supporting Information Table S2. For example, in the particular case of soil microbial communities, we gathered information for 101 plots including the 79 sites in Maestre et al. (2015) and the 22 sites in Delgado-Baquerizo et al. (2017).

Figure 1.

The global distribution of the 612 study sites used in this study. The map of Australia is enlarged to provide a better representation of the distribution of sites surveyed there.

2.2. Soil properties

For all 612 soil samples, we gathered information on organic C and pH. In all cases, soil organic C concentration was determined as described Anderson and Ingram (1993) and soil pH in a 1:5 soil-water extract with a pH meter. We also gathered information for 300 of these samples on total N and P. Soil total N was measured with a CN analyzer (LECO CHN628 Series, Leco Corporation, St Joseph, MI, USA). Total P was determined using a SKALAR San++ Analyzer (Skalar, Breda, The Netherlands) after digestion with sulphuric acid (3 h at 415°C) as described in Anderson and Ingram (1993). For 246 soils we assessed available N and P. Available N was assessed as the sum of NH₄⁺ and NO₃^-, following a 0.5 M NaHCO₃ (pH 8.5). Olsen inorganic P (Olsen et al., 1954) was measured following a 0.5M NaHCO₃ (pH 8.5) extraction following the soil-P fractionation protocol (Tiessen & Moir, 1993). These were measured colorimetrically (Sims et al., 1995) using a 0.5 M soil extracts and K₂SO₄ with a 1:5 soil: extract ratio (Delgado-Baquerizo et al., 2013). For 101 plots, we gathered information on the community composition, richness and abundance of soil fungi and bacteria (Supporting Information Table S1). These data were generated using next generation molecular analyses (Maestre et al., 2015). In brief, DNA was extracted from 0.5 g of defrosted soil samples using the Powersoil® Isolation Kit (Mo Bio Laboratories, Carlsbad, CA, USA). Quantitative PCR (qPCR) reactions were carried out in triplicate on an ABI 7300 Real-Time PCR (Applied Biosystems, Foster City, CA, USA).

Initial sequence processing and diversity analyses for both bacterial 16S rDNA and fungal ITS genes were conducted by using the QIIME package (Caprosa et al. 2019). Initially, low-quality regions (Q < 20) were trimmed from the 5′ end of sequences, and paired ends were joined with FLASH (Magoč & Salzberg 2011) for 16S rDNA sequences and Fastq-join (Aronesty 2011) for ITS reads. Sequences were demultiplexed, and a further round of quality control was conducted to remove sequences containing ambiguous bases (N) and reads containing bases with a quality score <25. Chimeric 16S rDNA sequences were detected by using the UCHIME algorithm from the USEARCH package (Edgar 2010, Edgar et al. 2011) implemented within VSEARCH (https://github.com/torognes/vsearch). The RDP training dataset (Version 9; Cole et al. 2005) was used as a reference for chimera detection, as recommended by the UCHIIME documentation. De novo (abundance-based) chimera detection was used for ITS data using USEARCH (Edgar 2010). The remaining high-quality chimera-free sequences were used for downstream analysis. Operational taxonomic units (OTUs) were defined as clusters of 97% sequence similarity using UCLUST (Edgar 2010). Taxonomy was assigned using UCLUST (Edgar 2010) against the Greengenes database (Version 13_850) for 16S rDNA OTUs (DeSantis et al. 2006, McDonald et al. 2012). For fungal ITS sequences, taxonomy was assigned by using BLAST (Altschul 1990) against the UNITE database (Version 6.9.7; Kõljalg 2013) (E < 10⁻⁵). This database, however, considers the Zygomycota as a phylum, whereas the most recent taxonomical references no longer do so (McLaughlin & Spatafora 2014). We thus refer to Zygomycetous fungi when referring to the different taxonomical units formerly included in the Zygomycota (McLaughlin & Spatafora 2014). The resultant OTU abundance tables for both primer sets were filtered to remove singletons and rarefied to an even number of sequences per samples to ensure an equal sampling depth (11,925 and 17,000 for 16S rDNA and ITS, respectively). Shannon diversity was calculated on these rarefied OTU tables by using QIIME (Caparoso et al. 2010). We estimated diversity using this metric because it has been recommend when quantifying and comparing microbial diversity (Haegeman et al. 2013). The number of bacterial sequences obtained from two of the sites surveyed was too low to estimate microbial diversity accurately, so they were not used in further analyses.

The total abundance of bacterial 16S-rRNA genes and fungal internal transcribed spacer (ITS) were amplified with the Eub 338-Eub 518 and ITS 1-5.8S primer sets (Evans & Wallenstein, 2012). The community composition and richness of fungi and bacteria were analysed in the Next Generation Genome Sequencing Facility of the Western Sydney University (Australia) using the Illumina MiSeq platform (Caporaso et al., 2012) and the 341F/805R (bacteria) and FITS7/ITS4 (fungi) primer sets (Herlemann et al., 2011; Ihrmark et al., 2012). After bioinformatics analysis (see Maestre et al., 2012 for methodological details), rarefied OTU (Operational Taxonomic Units) tables were used to calculate richness of OTUs and community composition of bacteria and fungi (as described in Maestre et al., 2015). Samples from these 101 plots were analysed together to ensure OTU-level comparability across samples.

2.3. Statistical analyses

Differences in soil chemical properties and relative abundance of microbes between Australia and Elsewhere were explored using a simple mixed model, based on a one-way ANOVA with subsamples. The contrast of interest, namely samples representing the Australian continent versus samples from countries in the rest of the world, was compared with an experimental error term formed from the stratum variance among 31 survey ‘units’. The survey units were the different countries, but the number of survey units (n = 31) was larger than the number of countries surveyed (n = 19) simply because multiple independent surveys were conducted in some countries. Therefore, in the case of samples from Argentina, there were three independent survey units, each carried out by a specific sampling team and research group. A survey ‘unit’ was considered to be independent if it were undertaken by a particular survey team within a specific area. Analyses were conducted using Genstat (VSN Internationa1 2015).

We supported these analyses with Structural Equation Modelling (SEM; Grace 2006) of the direct effects of two important climatic variables, mean annual temperature (MAT) and aridity on the five soil response variables (pH, C, total and available P, total N) and the three nitrogen and phosphorus ratios (C:N, C:P, NP). We also included in the model geographic location (distance from the Equator), and Australia (compared with elsewhere) and community type (grasslands vs woodlands). Our structural equation model is based on the plausibility of an a priori model explaining the relationships among a group of variables of interest (Appendix S1). The a priori model examined the direct and indirect effects of climate, distance from the Equator, community type, and Australia vs Elsewhere on our soil variables. We were particularly interested in whether the path coefficient between Australia and the variable of interest increased, reduced or had no effect on the variable of interest while excluding the effects of distance from the Equator, community type, temperature or aridity. Data on aridity, which expresses precipitation in relation to potential evapotranspiration, were calculated as 1-FAO Aridity Index using FAO’s global aridity map (http://ref.data.fao.org/). Mean annual temperature data were obtained from the Worldclim database (Flick & Hijmans, 2017).

The a priori model was compared with the variance-covariance matrix of our data to enable an overall goodness-of-fit to be assessed, using the χ² statistic. The goodness of fit test estimates the likelihood of the observed data given the a priori model structure. Thus, high probability values indicate that these models are highly plausible causal structures underlying the observed correlations. Analyses were performed using the AMOS 22 (IBM, Chicago, IL, USA) software. For each of our models, those with low χ², high Goodness of Fit Index [GFI] and high Normal Fit Index [NFI]) were interpreted as showing the best fit to our data (Appendix S2).

2.4. Microbial network analyses

These analyses were conducted using the two datasets including microbial data for fungi and bacteria (Global Network study and East Australia study). Bioinformatic analyses were done together for these two datasets, so that OTUs of fungi and bacteria are directly comparable (see details above). Using this data, we identified ecological clusters (or ‘modules’) of strongly associated taxa using correlation networks (‘co-occurrence networks’) and the following protocol. First, because of the enormous number of OTUs (synonymous with species) detected for fungi and bacteria, and to obtain a practicable network of interactions, we focused on the common species for these organisms (taxa accounting for 70% of the relative abundance of bacterial or fungal communities). These analyses were done independently for fungi and bacteria. These bacterial and fungal taxa were then merged into a single abundance table. This resulted in a dataset with 4192 taxa including 3608 bacterial and 584 fungal phylotypes. We then calculated all pairwise Spearman’s rank correlations (ρ) between all soil plant/animal and soil microbial/animal taxa. We focused exclusively on positive correlations because they provide information on microbial taxa that may respond similarly to environmental conditions (Barberán, Bates, O’Casamayor & Fierer, 2012). We considered a co-occurrence to be robust if the Spearman’s correlation coefficient was > 0.25 and P < 0.01 (see a similar approach Bastian, Heymann & Jacomy, 2009). The network was visualized with the interactive platform gephi (Bastian et al., 2009). Finally, we used default parameters from gephi to identify modules of soil taxa strongly interacting with each other.

We then computed the relative abundance of each module by averaging the standardized relative abundances (z-score) of the taxa that belong to each module. By standardizing our data, we ruled out any effect of merging data from bacteria and fungi. To further explore differences in microbial phylotypes (analogous to species) in dryland soils between Australia and elsewhere, we generated an ecological network of soil microbes using information from the microbial dataset (see above). We identified and calculated the relative abundance of six major ecological clusters (‘modules’) of phylotypes strongly co-occurring (Appendix S3).

Finally, we tested for differences in bacterial and fungal community composition between Australia and elsewhere using one-way permutational multivariate analysis of variance (PERMANOVA; Anderson 2001) on relatively abundant (> 70%) taxa. Non-metric multidimensional scaling ordination (nMDS) and the Bray-Curtis dissimilarity metric were used to explore overall differences in microbial composition (at the OTU level) between Australia and elsewhere. PERMANOVA and nMDS analyses were done using PRIMER-E Ltd. & PERMANOVA version 6 (Plymouth Marine Laboratory, UK).

3. Results

3.1. Soil chemistry

Australian soils had significantly lower pH values than elsewhere (F_1,29 = 8.52, P = 0.007; Fig. 2a). Averaged over all 612 sites, we found no differences in organic C between Australia and elsewhere (P = 0.94), but values were 40% lower in Australian soils when we restricted our analyses to grasslands and shrublands (F_1,24 = 5.40, P = 0.029; Fig. 2b). These results were maintained even when we accounted for differences in distance from the Equator, mean annual temperature and aridity using SEM (Fig. S4.1). Total (F_1,22 = 4.66, P = 0.043) and available (F_1,21 = 8.78, P = 0.007) P were significantly lower in Australian soils (Figs. 3a & 3b), but there were no differences in total N (P = 0.45, Fig. 3c). Ratios of soil C:N and C:P were significantly greater in Australian soils (F_1,23 > 12.2, P < 0.002; Figs. 3d & 3e), but the N:P ratio did not differ (P = 0.95; Fig. 3f). As with soil pH and organic C, our results were maintained after accounting for differences in distance from the Equator, mean annual temperature and aridity using SEM (Figs. S4.2 & S4.3).

Figure 2.

Mean (+ 95% CI) values of pH (a) and soil organic carbon (b) for sites in Australia and elsewhere. The upper panels show the mean values for Australia and elsewhere and the lower panels the values for different continents. * indicates a significant different in the mean values at P < 0.05 (One-way ANOVA). For organic C, the lower panel separates sites into grassland and woodland communities. Woodland communities were sampled only in Australia and Africa (see Supporting Information Appendix S1).

Figure 3.

Mean (95% CI) values of (a) total P, (b) available P, (c) total N, and the (d) C:N, (e) C:P, and (f) N:P ratios for Australia and elsewhere. The upper panels show the mean values for Australia and elsewhere and the lower panels the values for different continents. * indicates a significant different in the mean values at P < 0.05 (One-way ANOVA; see Supporting Information Table S2).

3.2. Microbial communities

Australian soils had lower bacterial richness (3108 cf. 3923; F_1,14 = 20.3, P < 0.001) but similar fungal richness (626 cf. 549, P = 0.19) to soils from global drylands (Fig. 4). The fungal to bacterial ratio was almost three-times lower in Australian soils (ratio: 0.05) than elsewhere (ratio: 0.14; F_1,11 = 6.06, P = 0.015; Fig. 4). When we partitioned these data between woodland and grassland sites, bacterial richness and the fungal to bacterial ratio were always less in Australia than elsewhere, but there were no community effects for fungal richness (Fig. 4).

Figure 4.

Mean (+ 95% CI) for the ratio of fungi to bacteria, bacterial richness and fungal richness for sites in Australia (n = 29) and elsewhere (n = 72) partitioned into grasslands and woodlands. Values differ significantly between woodland and grassland for all attributes except bacterial richness in Australia.

Proteobacteria (F_1,11 = 7.89, P = 0.013) was relatively more abundant, and Actinobacteria (F_1,11 = 5.84, P = 0.056) relatively less abundant, in Australian soils (Appendix S6). For fungi, Australian soils had a greater relative abundance of Basidiomycota (29.3 cf. 19.5; F_1,11 = 15.58, P < 0.001), but lower relative abundance of Glomeromycota (0.37 cf. 3.92; F_1,11 = 11.42, P = 0.002) than elsewhere. When we correlated microbial relative abundance with soil chemical data, we found that Proteobacteria was strongly correlated with low pH and low total P (Appendix S5). Conversely, abundance of Gemmatimonadetes and Verrucomicrobia was positively associated with soil pH, total and total and available P. For fungi, Basidiomycota were relatively more abundant (F_1,100 = 14.07, P < 0.001) and Chytridiomycota less abundant in Australian soils (Appendix S6).

Our microbial data clustered into six clear networks (Fig. 5a). The relative abundance of Modules #0 and #1 was significantly lower for Australian soils than elsewhere (Fig 5b), and positively correlated with increasing pH and soil P (Appendix S5). Modules #3 and #4, however, displayed the opposite trend, being negatively correlated with pH and soil P and relatively more abundant in Australian soils (Appendix S5). Specific taxa comprising each module and including multiple mycorrhiza, saprobe and pathogen species and their traits are available in Appendix S3. The nMDS analyses indicated clear and significant separation between both bacteria and fungi (Fig. 5c).

Figure 5.

Correlation network (a) including multiple bacterial and fungal phylotypes strongly co-occurring with each other within microbial modules, (b) their relative abundance, and (c) biplots of the first and second axes of the 2D NMDS solution for Bacteria and Fungi. Significance levels as follow: **P < 0.01.

4. Discussion

Compared with elsewhere, two of the strongest characteristics of Australian soils were their lower soil P and pH. Australian soils are known to be among the oldest soils on Earth, a consequence of the minor effect of last glaciations on their soils (McKenzie, Jacquier, Isbell & Brown, 2004). Long-term chronosequence studies (Vitousek et al., 2002) and studies of ancient tropical systems (Delgado-Baquerizo et al., 2016) demonstrate that older soils have extremely low levels of P. This phenomenon is also apparent in drylands, as evidenced by the classic tudies of Lajtha (1988) and Lajtha and Schlesinger (1988) in the Chihuahuan Desert in the western United States. Phosphorus is largely under abiotic control, and derived mainly from subsoil and P-rich parent material (Vitousek et al., 2010). However, P can also be deposited through low-distance aeolian deposition (Das, Evan & Lawrence, 2013). Lower total and available P is consistent with the large number of studies of plant-soil relationships in Australia (see reviews by Lambers, Raven, Shaver & Smith, 2008; Lambers, Brundrett, Raven & Hooper, 2010), though P levels are spatially variable (Kooyman, Laffan & Westoby, 2017). Similarly, chronosequence studies reveal that soil pH tends to be higher in young soils and declines with soil age (Alfaro et al., 2017).

The pattern of pH change in Australia is strongly controlled by precipitation and leaching (high pH in drylands), bedrock characteristics (e.g. high over calcrete, Kooyman et al. 2017) and vegetation (low organic matter in drylands; de Caritat, Cooper & Wilford, 2011). Soil pH, which was also lower in Australian soils, typically declines with ecosystem development due to prolonged leaching of cations in the bedrock over millennia. However, the extent to which pH changes with age depends on soil type and particularly, the nature of the parent material (Lambers, Shane, Cramer, Pearse & Veneklaas, 2006). Although the effect of low P on terrestrial biota is clear from the Australian literature (McKenzie, Jacquier, Isbell, & Brown, 2004; Lambers et al., 2006; Lambers et al., 2010), we lack a comprehensive comparison of the magnitude of the differences between Australia and other continents, particularly from drylands (Stafford Smith & Morton, 1990). Interestingly, many plant species from Australian drylands (e.g. family Proteaceae) have co-evolved under conditions of low soil P (Specht & Rundel, 1990) by developing proteoid or clustered root structures, or by forming close associations with N-fixing microbes or mycorrhizal fungi. These root morphologies allow plants to exude large amounts of P-mobilizing carboxylates (organic anions), making then extremely competitive in P-depleted environments (Lambers et al., 2006). Similarly, long-lived structures on plants may allow those growing on infertile soils to produce C relatively more efficiently than N, allowing Australian drylands to produce relatively high levels of biomass, even in the presence of low levels of P (Orians & Milewski, 2007). These vegetation communities, which are extremely well adapted to low P, occur widely across the Australia drylands, further supporting the notion that Australian soils have much lower levels of P than drylands elsewhere.

4.1. Soil C levels in Australian soils are similar to soils elsewhere

Given the expectation that Australian soils should reflect characteristics of retrogressive ecosystems, we had expected their soils to have less C than global drylands elsewhere. We partially fulfilled our expectation, as soil C was lower in Australia than elsewhere, but only when forests and woodlands were excluded from our analyses. Eucalypt woodlands are a widespread vegetation community in Australian drylands and may explain the higher C:N ratio we found in Australian soils, consistent with studies of major forested biomes worldwide (Xu et al., 2013). Several reasons may support the lack of an overall difference between soils in Australia and elsewhere. First, drylands soils are naturally low in C (Plaza, Gascó, Méndez, Zaccone & Maestre, 2018). The low capacity of dryland soils to retain C, which has recently been highlighted by Rabbi et al., (2015) is a characteristic that seems to be shared across drylands worldwide. Second, large areas of Australia’s drylands support dense forests and woodlands, often dominated by Eucalyptus spp., rather than low-stature shrubs and grasses. This is largely because many of these trees are able to access groundwater (Eberbach, 2003), releasing them from a reliance on precipitation or the redistribution of surface water. Soil C storage is known to be significantly greater beneath woody communities than grasslands (Chen, Hutley & Eamus, 2005), a key condition that might have helped Australian soils maintain higher than expected levels of C, given their long history of development. For example, soil organic carbon stocks in northern Australian savannah woodlands are known to be almost three-times greater than global averages, and similar to tropical woodlands (Chen et al., 2005). Dense woodlands and forests also have a greater capacity to develop on nutrient-poor soils (Bond, 2010). More interestingly, when we re-analysed our data excluding sites with forests, woodlands and savannas, Australian soils had about 40% less C than dryland soils elsewhere. This is in agreement with the expectation that ecosystems in Australia might be largely retrogressive, as defined in Peltzer et al., 2010). The trajectory of soil C change during ecosystem development is non-linear (quadratic) with time. As ecosystems develop, C production increases rapidly via microbial processes, but begins to decline in very old systems as rates of nutrient leaching exceed accumulation and old soils are no longer capable to retain organic matter, in part, a consequence of the largely leached soil cations over millennia (Vitousek et al., 2010). Interestingly, unlike total C and P, total N in Australia did not differ from elsewhere. Drylands are known to have a very active N fixer (e.g. cyanobacteria) and nitrifier microbial communities (Delgado-Baquerizo et al., 2016). Moreover, unlike P, which is linked to the bedrock availability, the major source of N is the atmosphere, explaining why Australian drylands have maintain similar levels of total N and N availability than drylands elsewhere.

4.2. Australian soils have a unique microbial signature

Differences in aridity, soil mineralogy and geomorphology are known to be strong drivers of surface- and subsurface-resident microbial communities (Pointing & Belnap, 2012). At the entire community composition and phylum levels, we found strong and significant differences between soils from Australia and elsewhere. For example, the NMDS plots of bacterial and fungal communities showed a significant spatial separation across locations from Australia vs. Elsewhere. Also, important phyla such as Proteobacteria, the dominant bacterial taxon globally (Ramirez, Craine & Fierer, 2012; Maestre et al., 2015; Delgado-Baquerizo et al., 2016), had a greater relative abundance in Australian soils than elsewhere. A higher amount of Proteobacteria has been reported previously during long-term ecosystem development in soils from the Southern Hemisphere (Jangid, Whitman, Condron, Turner & Williams, 2013).

Some of these general patterns might be related to the low P and pH that characterize soils from Australia compared with elsewhere. For example, soil pH values from our soils were also negatively correlated with the relative abundance of Basidiomycota (Supporting Information Table S4), a group of fungi found typically in acidic forested soils, but positively correlated with Actibobacteria and Gemmatimonadetes, and the fungal phyla Glomeromycota and Chytridiomycota (Supporting Information Table S4). Proteobacteria, particularly γ-Proteobacteria, have been shown to be important drivers of soil functioning in both field and microcosm studies from Australian soils (Delgado-Baquerizo et al., 2017).

Our network analyses further indicated that some dominant taxa (accounting for >70% relative abundance of all taxa) co-occur in Australia but are not as abundant elsewhere. Our results suggest therefore that the specific soil properties from Australian soils likely promote particular ecological clusters of strongly co-occurring microbial species, clusters that are different elsewhere. For instance, soils from Australia have very low relative abundance of Modules 0 and 1, which however, are more abundant elsewhere (Fig. 4). These differences could have important implications for soil functioning, given the role of some of microbial taxa comprising the modules, on processes such as the production of enzymes to decompose soil organic matter (i.e. starch and cellulose degradation), the release C for maintaining structural components, or the degradation of chitin, to name a few (Trivedi, Delgado-Baquerizo, Anderson & Singh, 2016). Modules 0 and 1, whose members were significantly less abundant in Australian soils (Fig. 5a), included phyla that are positively related to soil pH (Actinobacteria, Glomeromycota) and total and available P (Bacteroidetes; Table S4). In contrast, Clusters 3 and 4 contained Proteobacteria, a common bacterial phylum in Australian soils, which is negatively associated with soil pH. Differences soil P and pH are known drivers of microbial communities in terrestrial ecosystems (Lauber et al. 2009), and there is strong evidence that these differences are linked to the assembly of microbial communities in Australia (Lambers et al. 2008).

Conclusions

Together, our study provides evidence that Australian dryland soils are different from drylands elsewhere in terms of their main properties and microbial communities. The signature of Australian soils was consistent with that of highly weathered landscapes (Vitousek et al., 2010; Wardle et al., 2014) low in pH, lower contents of total and available P, and generally higher C:N and C:P ratios, a greater relative proportion of Proteobacteria and lower abundance of Actinobacteria and, more importantly, higher relative abundance of specific microbial assemblies linked to low soil pH. It is critically important that we understand how Australian soils differ from those of other drylands because it will shield us against unrealistic expectations about the extent to which current land management practices such as conservation farming or low risk stocking might lead to better environmental outcomes for them. Second, knowledge about Australian soils will help to explain why, for example, Australian ecosystems and their biota behave in idiosyncratic ways; adding to the growing body of knowledge and attempts to evaluate whether Australian dryland soils are indeed resource poor, or whether this notion is an artificial construct arising from a Eurocentric view of ecosystem productivity. This knowledge will ultimately provide us with a greater understanding of the mechanisms underpinning the processes shaping the Australian environment and its biota.

Biosketch

David J Eldridge is a dryland ecologist with the NSW Office of Environment and Heritage based at the University of NSW. The focus of his work is the management of drylands, in particular, how grazing by different vertebrate herbivores, soil disturbing animals, shrub encroachment and biological soil crusts influence soil and ecological processes.