The known and the unknown in soil microbial ecology

Abstract

The methodical developments in the fields of molecular biology and analytical chemistry significantly increased the level of detail that we achieve when exploring soils and their microbial inhabitants. High-resolution description of microbial communities, detection of taxa with minor abundances, screening of gene expression or the detailed characterization of metabolomes are nowadays technically feasible. Despite all of this, our understanding of soil is limited in many ways. The imperfect tools to describe microbial communities and limited possibilities to assign traits to community members make it difficult to link microbes to functions. Also the analysis of processes exemplified by enzyme activity measurements is still imperfect. In the future, it is important to look at soil at a finer detail to obtain a better picture on the properties of individual microbes, their in situ interactions, metabolic rates and activity at a scale relevant to individual microbes. Scaling up is needed as well to get answers at ecosystem or biome levels and to enable global modelling. The recent development of novel tools including metabolomics, identification of genomes in metagenomics sequencing datasets or collection of trait data have the potential to bring soil ecology further. It will, however, always remain a highly demanding scientific discipline.

soil ecology, bacteria, fungi, enzyme activity, community composition, ecosystem processes

INTRODUCTION

The importance of soil ecology increased dramatically with the appreciation of soil as a major component of the global carbon cycle and the fact that the ongoing climate change may alter C balance and make global soils either sinks or sources of carbon dioxide or other important greenhouse gases (Crowther et al. 2016). In the past, soil was considered to be a black box, accessible only to the monitoring of elemental composition, gas fluxes or total microbial biomass and soil microbiota was almost exclusively characterised using isolation and culturing techniques. Following the period of soil DNA research, using fingerprinting and cloning, the development of high-throughput nucleic acid sequencing and fine analytical methods fortunately increased our ability to describe soil at a fine resolution. This resolution at the moment goes as far as to the description of microbial community composition at genus or subgenus level and assessing the functional potential and its expression on the level of individual genes. As an example, it is possible to use soil metagenomics and metatranscriptomics to describe the community-wide carbon use (Žifčáková et al. 2017) or N cycling (Hesse et al. 2015; Mackelprang et al. 2018) covering soil bacteria, fungi and other members of the soil biota. Comprehensive use of multi-omics methods give us some hint about the genomic potential of microorganisms in various soil and the involvement of individual taxa in soil processes (Hultman et al. 2015; Woodcroft et al. 2018).

Still, in my opinion, we are far from the perfect understanding of soil functioning. Many papers heavily rely on the description of microbial communities as if these can represent a key to the understanding of soil functions. Although functional aspects of soil functioning, such as the analysis of biochemical processes in soils, are also frequently addressed, current methods are often not simple to use and their results are difficult to interpret. Finally, the view of the soil is often offered from only one perspective, the perspective of the soil sample recovered. There are, however, many other, possibly more relevant perspectives of soil composition and functioning: the perspective of individual taxa and their activity, the identity of soil microenvironment (such as a soil pore or aggregate) or the ecosystem or biome-level perspective that may bring us up to the understanding of Earth functioning. All of these seem to be too difficult at the moment.

The aim of this review is to indicate the weak points in the mainstream approaches in the soil microbial ecology research and to offer some ideas where to focus our efforts to move the understanding of soil functioning forward.

Microbial communities: going beyond taxonomic lists?

The description of the composition of bacterial or fungal communities in soils seems to be a key component of most papers devoted to soil microbial ecology. It is not surprising, if we consider that the high-throughput sequencing approaches that made these analyses possible are available for less than the 10 last years and their application undoubtedly made our picture of the composition of soil microbiota much more complete (Tedersoo et al. 2014; Thompson et al. 2017; Delgado-Baquerizo et al. 2018).

Although the resolution and coverage of molecular markers targeting bacterial 16S within the rRNA operon that are currently used seem to be reasonable (Caporaso et al. 2012), the picture of the bacterial community that we obtain is still very often imperfect. Even if the coverage of common primers is fine, many groups of bacteria may be missing from amplicon-based surveys and only emerge when total metagenomic DNA is sequenced (Hug et al. 2016). More importantly, it is often forgotten that 16S sequence abundances do not represent taxon abundances and these terms are frequently mixed. This is caused by the fact that bacteria possess variable numbers of 16S rRNA genes in their genomes, ranging from 1 to 16, a phenomenon already recognised long time ago (Farrelly, Rainey and Stackebrandt 1995). The reason for this variation has been a subject of some debate (Klappenbach, Dunbar and Schmidt 2000) which is, however, inconclusive and does not affect the fact that sequence counts do not correspond to cell counts as obtained, for example, using cell sorting after FISH staining. Fortunately, there are databases available that list the known genomes of bacteria and the numbers of rRNA operons per genome (Stoddard et al. 2015) and methods have been proposed how to obtain genome abundance estimates equivalent to cell abundance estimates from 16S sequencing data (Větrovský and Baldrian 2013; Angly et al. 2014). Typically, these correction methods are based on the observation that closely related bacteria have similar rRNA operon numbers and the rRNA copy number in the closest known bacterial genome offers a correction factor for calculation of genome copies from 16S copies and by definition, they are not equally precise for all taxa (Louca, Doebeli and Parfrey 2018). The relative composition of soil bacterial communities after correction may be quite distant from the composition of the 16S sequence pool, where bacteria with more rRNA copies are overrepresented (Větrovský and Baldrian 2013). It should be also noted that in some cases, single bacterial genome contains different 16S rRNA genes that show low sequence similarity and in silico clustering methods recognise them as different species (Větrovský and Baldrian 2013). Even though this phenomenon is rather rare, it may affect bacterial diversity estimates.

Finding the optimal molecular marker for fungal community sequencing is much less easy. While ITS seems to be the primary marker for fungi due to the fact that it offers the best taxonomic resolution (Schoch et al. 2012), its use has multiple limitations. Importantly, ITS cannot be aligned across distant taxa and thus cannot be used for phylogenetic studies. This is why there was a consistent struggle to develop fungal marker based on 18S (SSU) or 28S (LSU) rRNA regions. Unfortunately, even promising candidate regions have too low taxonomic resolution or are too long for high-throughput sequencing (Porras-Alfaro et al. 2014). Also, the whole ITS, covering both ITS1 and ITS2 regions, is too long for HTS and one of them has to be selected. ITS regions of different fungi have very different lengths, which represents another problem: in PCR amplification, short sequences are typically more frequently amplified, and this may result in biased community representation. The current recommendations prioritize ITS2 (Ihrmark et al. 2012) since it is more consistent than ITS1 (Walters et al. 2016) in terms of size variation and >75% of fungal amplicon data generated today use this region. Even the best performing primers for fungal ITS2, however, show biases in recovery of various fungal groups, often prioritizing Dikarya (Ihrmark et al. 2012; Větrovský et al. 2016). Due to a multitude of limitations, some groups recommend to use sequencing of longer amplicons for fungi that improve coverage of the tree of life and reduce size variation (Tedersoo, Tooming-Klunderud and Anslan 2018). These markers, however, need to be sequenced on a PacBio platform, which is considerably more expensive than Illumina sequencing (Nilsson et al. 2019).

Any marker targeting fungal rRNA (LSU, SSU, and ITS alike), however, also suffers from the variation of copy numbers per fungal genome. The copy counts are estimated to range from a single copy to hundreds per fungal genome (Raidl, Bonfigli and Agerer 2005; Debode et al. 2009; Herrera et al. 2009), but are unknown for the vast majority of fungal species since their presence in a repetitive cassette does not allow their reliable quantification. Unfortunately, the copy number also does not seem to reflect the size of fungal genomes (Baldrian et al. 2013,b).

The problems with unknown copy numbers of rRNA in fungal genomes could potentially be solved by choosing a single copy gene as a molecular marker. However, as with any protein marker, the choice of PCR primers is difficult. Because of the degeneration of genetic code and most primer pairs are either nonspecific or have a limited coverage. To certain degree, the use of the rpb2 gene as a molecular marker showed sufficient discriminative power and reasonable coverage of the fungal tree of life. The comparison of ITS2 and rpb2-based primers showed very different results of fungal community composition, especially for the representation of basal fungal lineages that was low with the former primer and high with the latter one (Větrovský et al. 2016). The results also indicated that ITS sequences in the same fungal genome often show highly divergent sequences, which may inflate diversity estimates (Lindner and Banik 2011). With the rpb2 marker, these problems are reduced (Větrovský et al. 2016).

If the choice of PCR primers is difficult for fungi, it is virtually impossible for highly divergent taxa, such as soil protists where general primers do not exist and any primer-based approaches are limited to lower taxonomic groups (Geisen et al. 2015a). For obvious reasons, viral diversity cannot be addressed by PCR-based approaches. Instead, current combinations rely on data mining in metagenomic and metatranscriptomic data with or without the use of isolation of virus-specific particles or nucleic acids (Paez-Espino et al. 2016; Pratama and van Elsas 2018; Trubl et al. 2018).

The suitable alternative to amplicon-based studies that are biased by primer limitations and copy number issues may be the sequencing of RNA. Total RNA contains mostly ribosomal RNA, molecules that typically carry sufficient taxonomic information even within short reads. Thus, after sequencing, RNA reads corresponding to LSU or SSU may be identified and assigned taxonomy e.g. by comparison to the RDP or Silva databases (Quast et al. 2013; Cole et al. 2014). The read count reflects the ribosome pool of taxa covering the whole tree of life and the ribosome abundance appears to be a reasonable proxy of relative abundance of organisms. Although community analysis using the high-throughput sequencing of total RNA was already demonstrated long time ago (Urich et al. 2008), the method is not widely applied. It was, however, demonstrated to cover a large variety of organisms in the understudied arctic peatlands (Tveit, Urich and Svenning 2014), and successfully used to demonstrate the width of protist diversity in soils (Geisen et al. 2015b). The slightly lower taxonomic resolution and complications of RNA work are, however, the downsides of its application.

The community surveys offer a picture of composition of the soil microbiome that may be highly relevant. However, it needs to be noted that soil microbial communities often show limited change in time (Žifčáková et al. 2016), partly due to limited activity or dormancy of certain taxa (Lennon and Jones 2011). Even if the microorganisms are active, the community composition is likely similar across time if soil properties remain constant. Since we know that the growth rates of soil bacteria differ tremendously in culture (Lladó et al. 2016) as well as in soil (Goldfarb et al. 2011), the steady state has to be maintained by turnover (growth and death) that is fast in some taxa and slow in others. The fraction of C and nutrient flow through the fast-growing bacteria must be much higher than through the slow growing or dormant ones. To estimate microbial contribution to ecosystem nutrient cycling and other processes, the community composition data thus need to consider in situ growth rates of individual taxa. Fortunately, it seems that methods to make such growth rate estimates start to appear, e.g. the study of the rate of incorporation of ¹⁸O-labelled H₂O into DNA (Hungate et al. 2015).

When performing DNA- or RNA-dependent studies, one should not forget about the extraction bias that represents another factor of uncertainty (Sagova-Mareckova et al. 2008; Albertsen et al. 2015). The optimization of extraction procedures considering the specific aims of study is always required.

Linking microbial taxa, ecosystem functions and traits

If the only use of the microbial community composition would be the information who is abundant under certain conditions or how taxa contribute to C flow, there would be little use of such information since the important questions why are they abundant or what they do would remain unanswered. In fact, community sequencing is typically performed with the expectation that taxonomic affiliation may bring some functional information about the soil microbiome. This is theoretically possible, if functional traits can be assigned to higher or lower rank microbial taxa and the identification of the taxon can thus relate to function. Indeed, some studies show that at least some traits are conserved at various taxonomic levels, although many others seem to be only common to very closely related taxa (Medini et al. 2005; Martiny, Treseder and Pusch 2013; Goberna and Verdu 2016). This should be also true for important ecological traits (Zimmerman, Martiny and Allison 2013) and the classification of sequences of environmental amplicons into molecular taxa—typically called operational taxonomic units (OTU) is motivated by the expectation that they represent ecologically homogeneous groups and their abundance in soil will be informative considering their ecosystem role.

Taken this reasoning further, the assumption that genomes of closely related taxa are functionally similar resulted in the development of prediction tools that try to infer genome content from 16S amplicon sequences by taking the most closely related genome as a proxy, for example PICRUSt and Tax4Fun (Langille et al. 2013; Asshauer et al. 2015). These tools are being applied in various context including soil (Oh et al. 2016; Li et al. 2017). However, recent meta-analysis shows that the assumption of genome similarity is not valid even within bacteria showing 97–99% 16S sequence similarity that belong to the same OUT. Genomes of such bacteria have been demonstrated to share as much as 95% genes, but also as few as 25% (Lladó, Větrovský and Baldrian 2019). The differences in gene content are not restricted to dispensable genes: even for the highly relevant gene families of glycoside hydrolases, essential for the utilization of various C sources, 40% of gene families are present in one genome of the OTU while absent in another one, i.e. the ecology of C utilization differs within the OTU (Lladó, Větrovský and Baldrian 2019). It is thus clear that genome content and their functional predictions based on 16S similarity are unreliable.

Since the assignment of traits and functions is essential and the comparison to genome databases does not seem reliable, there is a question how to assign functional traits to microbes. The most obvious method is the characterization of isolates. After the arrival of DNA-based methods, bacterial isolation was often overlooked, although the improvement of culturing techniques such as long-time culturing, in situ cultivation or dilution of media increased our ability to isolate so far uncultured taxa (Davis, Joseph and Janssen 2005; Pankratov et al. 2008; George et al. 2011; Stewart 2012). It was demonstrated that a substantial share of the dominant bacterial taxa can be isolated using these tools and subsequently characterised (VanInsberghe et al. 2013; Lladó et al. 2016) and also many dominant soil saprotrophic fungi, including yeasts can be isolated and characterized (Masinova et al. 2018; Yurkov 2018).

The isolation efforts that offer the possibility of comprehensive characterisation of isolates including genome sequencing, however, are technically demanding, and allow the characterisation of a limited number of taxa. For certain traits, such as the screening of substrate utilization, strain characterization can be performed with much higher throughput by sequencing of DNA amplicons labelled with stable isotopes (Eichorst and Kuske 2012; Štursová et al. 2012; López-Mondéjar et al. 2018), or simply by sequence-based identification of those associated with the substrate of interest, e.g. fungal biomass or cellulose (Brabcová et al. 2016; Bhatnagar, Peay and Treseder 2018). EpicPCR, the method that allows to co-amplify two genes distantly located in the same genome, represents an inventive technique that allows to link the occurrence of 16S tag with the sequence of another gene of interest (Spencer et al. 2016) and will be hopefully used much more frequently in the future. Potentially, the data mining in the existing literature, such as the phenotype records in species description may also represent a viable approach (Barberán et al. 2017).

Recently, specific tools have been developed for the identification of metagenomic contigs originating from the same genome and their binning into the so-called metagenome-assembled genomes—MAGs (Boisvert et al. 2012; Li et al. 2015; Nurk et al. 2017; Parks et al. 2017). If metagenome sequencing is deep enough, tens to hundreds of MAGs may be recovered from soil samples and functionally characterized by screening their gene content (Taş et al. 2018; Woodcroft et al. 2018). Both MAG completeness and quality (presence of contaminant sequences) vary and there is a question if they represent biological reality rather than chimeric constructs (Bowers et al. 2018), but still they seem to be very promising proxies of bacterial taxa (construction of fungal MAGs is so far unfeasible). Single-cell genomes may represent another tool to characterize the genome potential of microbial taxa, the fractions of genomes recovered from soil microbial cells is, however, typically very low and this largely limits the applicability of this tool (Bowers et al. 2018). Hopefully, creative way to link microorganisms or their proxies—rRNA sequences—to functional traits is one of the grand challenges of today's microbial ecology.

Soil process rates and how to link them to microbes?

To understand soil processes, it is not enough to characterize microbial community composition, their potential and traits. It is essential to measure rates of soil processes and provide their link to microbes. In many cases, it may be enough to estimate the total microbial biomass or the biomass of higher taxa of soil organisms, since many soil processes scale up with microbial abundance (Kuzyakov and Blagodatskaya 2015). Estimation of total microbial carbon or phospholipid fatty acids may be convenient tools to get the total abundance of soil organisms. There are several tools that can be used to look for taxon specific abundances: PLFA or quantitative PCR are typically used to estimate the size and share of bacterial and eukaryotic (fungal) microorganisms in soils (Rousk et al. 2010; Frostegård, Tunlid and Bååth 2011). In the case of qPCR, it must be noted that the results are equally biased by rRNA copy number variation across genomes as is the community composition analysis. Specific methods may help to address the abundance of certain microbial groups, such as fungi (Baldrian et al. 2013b; Wallander et al. 2013) and protists (Geisen and Bonkowski 2018). Although biomass quantification methods seem to be technically feasible, use of different methods may still offer different estimates, as shown on the example of qPCR, PLFA and ergosterol to analyse fungal biomass in temperate forest topsoil (Baldrian et al. 2013).

The other component of the microbial process analysis is the measurement of soil process rates, another challenging issue. Enzyme assays were always considered attractive to soil ecologists, theoretically offering a direct link between soil organisms and biochemically well-defined reactions. Enzyme assays are indeed often used but unfortunately, they are often used improperly or address inappropriate targets (Nannipieri, Trasar-Cepeda and Dick 2018).

In the traditional view, cellulose decomposition was seen as a concerted action of three groups of enzymes: internal cleavage of cellulose polymers by endocellulases, liberating cellobiose disaccharides from chain ends by exocellulases and cellobiose cleavage by β-glucosidases (Lynd et al. 2002). Assays for all of these enzymes are available. However, recent research led to the discovery of alternative pathways of cellulose degradation: polysaccharide mono-oxygenases are highly transcribed on cellulose, and reported among the most frequent carbohydrate-active enzymes in soil metatranscriptomes (Hesse et al. 2015; Žifčáková et al. 2017), but the assay of their activity is not available. The same is true for the cleavage of cellulose and other polysaccharides via redox reactions related to the Fenton chemistry (Baldrian and Valášková 2008). Proteases represent another group of enzymes that are highly expressed, but difficult to assay due to the complexity of reaction mechanisms. On the other hand, urease activity is frequently reported and believed to represent organic N mineralization in general, while in fact urea is rare in most soils and represents just a very small part of soil organic N; the same is true for arylsulfatase and S cycling (Nannipieri, Trasar-Cepeda and Dick 2018).

While those enzymes whose reaction products can be detected in the presence of soil components can be measured in the context of soil (in the soil homogenate—'soil slurry'), others, typically endo-cleaving hydrolases or oxidative enzymes (laccases, peroxidases) have to be extracted before analysis because the corresponding reaction has to be followed under defined conditions in vitro (Baldrian 2009; Heinonsalo et al. 2012). The extraction of enzymes, however, recovers only the free fraction of enzymes while the fraction of enzymes bound to microbial cells, soil organic matter or minerals is difficult to assess. Estimates tell that it may account for as little as a few percents of the total activity (Štursová and Baldrian 2011; Mašínová, Yurkov and Baldrian 2018).

It needs to be noted that even in soil homogenates, assays measure potential, not actual in situ activity (Burns 1982). Although some assays are designed to mimic soil conditions like pH or temperature (German et al. 2011), in practice this is impossible to achieve (Lessard et al. 2013) since any additions to the soil matrix change at least the spatial context, water content and substrate availability. This is why it might be best to quantify enzyme molecules by using defined assay conditions (pH, temperature, substrate concentration) optimal for enzyme assays across samples and studies (Baldrian 2009). Reaction rates of soil enzymes have been shown to be highly temperature-dependent (Baldrian et al. 2013a; Bárta et al. 2014), so without information on this dependence and in situ temperature records, there is no way to estimate turnover rates.

Enzymes show considerable stability in time: when lifespan of enzymes in soils after fumigation was compared across multiple environments, the fraction of enzyme activity differed, but in many cases more than 50% of initial activity was still present after 12 weeks (Schimel, Becerra and Blankinship 2017). The stability of enzymes and other proteins may even reach over one year (Schneider et al. 2012), and it is clear that the protection of costly produced extracellular enzymes from degradation, e.g. by means of extensive glycosylation, reduces the costs of their production. Enzymes thus rather represent the situation in soil integrating a long period of time and are unfit to reflect dynamic changes: the changes in microbial community composition are typically much faster than revealed by enzyme activity measurements (Štursová et al. 2014; Kohout et al. 2018).

Apparent kinetic constants can be determined for soil enzyme extracts that describe the dependence of enzyme reaction rates on substrate concentrations (Eichlerová, Šnajdr and Baldrian 2012; Triebwasser-Freese et al. 2015), although these are affected by a wide variety of factors, including temperature (Allison et al. 2018). However, even if kinetic parameters of soil enzymes are known, without the knowledge of substrate concentration and its accessibility in soil, it is impossible to calculate actual reaction rates. Substrate limitation is probably very important in soils, especially in those with low organic matter content.

Metabolomics may theoretically provide the missing information about substrate availability in soil and provide the link between potential enzyme activities and real in situ process rates. In the last years, metabolomics is being increasingly applied to soil (Swenson et al. 2015; van Dam and Bouwmeester 2016; Petriacq et al. 2017), giving not only the estimates of soil nutrient pools, but also the opportunity to reconstruct pathways of metabolic dependence among soil bacteria (Swenson et al. 2018). Metabolome analysis typically identifies soil metabolites and shows the size of their pools at the time of sampling. However, maybe even more important than pools would be the estimation of metabolite fluxes, because the pools are dynamic and just represent the steady state difference between production and consumption. Although large pools may indicate high metabolite availability, they do not indicate fast turnover: phenolic compounds, for example, may be abundant in soils due to their recalcitrance or toxicity. On the other hand, compounds that are rapidly produced and rapidly degraded may have relatively small concentrations. In the future, when analytical methods allow it, transformation of labelled metabolites may potentially replace enzymes as a proxy of metabolic rates having the big advantage that they would represent the real reaction substrates in the real context.

Although not suitable to estimate biogeochemical rates, metagenomic and metatranscriptomic approaches seem to represent a feasible way to link microbial taxa with ecosystem processes. While metagenome sequencing following after labelling of xenobiotic utilizers by substrate-induced gene expression (SIGEX) allows to identify potential biochemical pathways of degradation (Meier, Paterson and Lambert 2016). The presence of certain genes in soil metatranscriptomes clearly indicates the need of their producers to perform the encoded biochemical reaction (Žifčáková et al. 2016). In addition to analysing whole metatranscriptomes, genes encoding specific functions are sometimes addressed using gene-specific PCR or RT-PCR methods (Chen et al. 2007; Kellner, Zak and Vandenbol 2010). The big disadvantage of these method is the difficulty to design primers with sufficient specificity and coverage for the same reasons mentioned above for the fungal protein-encoding genes. Without whole metatranscriptome analysis, it is unclear, how many genes with relevant function are missed from amplification and without amplicon sequencing, it is unclear what is the share of nonspecific sequences. Metaproteomics can give similar answers, but its use in complex environments is still challenging due to an array of technical problems ranging from extraction to annotation (Schneider et al. 2012; Bastida and Jehmlich 2016; Callister et al. 2018).

Considering soil complexity: scales, spatial heterogeneity and temporal dynamics

Although the methodical limitations of current soil microbial ecology outlined above represent per se a challenge to our exploration of soil functioning, we should not forget that the complexity of soil as a system itself brings multiple challenges that go beyond basic methodologies. Most importantly, it is the complexity of soil structure with vertical stratification, variation in space and the existence of specific microhabitats (Lindahl et al. 2007; Kuzyakov and Blagodatskaya 2015) as well as the dynamics of soil processes at multiple scales ranging from hours to centuries (Baldrian 2017). In most cases, sample analysis represents a picture of soil functioning at the time of sampling on the level of a soil core or a few soil cores. We should not forget that most bacteria have limited motility that is determined by habitat connectivity, such as the water content of soil (Wolf et al. 2013). Many bacteria, especially in dry soils, thus live on a single aggregate or in a single pore (Vos et al. 2013). Aggregate classes and probably individual aggregates themselves differ in organic matter content and microbial community composition (Davinic et al. 2012) and the co-occurrence of bacteria within soil core-size communities does not exist in reality. Soil microhabitats inhabited by microbes, such as plant root surfaces, rhizosphere or drilosphere often differ largely in properties, nutrient content, rates of biochemical processes and temporal dynamics (Baldrian 2014; Kuzyakov and Blagodatskaya 2015; Banfield et al. 2018; Zhalnina et al. 2018). Addressing individual microhabitats and perhaps scaling down to individual aggregates or pores is one of the ways to understand real conditions of microbial life, to identify activity hotspots and ‘coldspots’ and to understand interspecific interactions. In this sense, it is also important to note that most of the current soil research is restricted to the uppermost soil layers. Subsoil represents a specific environment with larger differences between activity hotspots and the soil matrix (Uksa et al. 2015; Heitkotter and Marschner 2018) that for sure requires more attention.

Apparently, the size of a soil core is also of limited use when we need to provide answers on the stand level, ecosystem level or biome level (Ekblad et al. 2013; Crowther et al. 2016). The scaling up to get representative data for a stand level is still feasible (Bahram, Peay and Tedersoo 2015; Štursova et al. 2016), but the larger the addressed area, the more difficult it is to feed models with reasonably precise predictions (Ekblad et al. 2013; Hernandez et al. 2017). Upscaling of process rates, pool estimates or microbial community data is probably one of the biggest challenges of the future soil microbial ecology. Scaling up also applies to the temporal aspect of ecosystem functioning where we should better appreciate the importance of temporal variation of ecosystem processes driven by temperature, moisture or other factors.

Outlook

How to move forward? We need to think about soil microbial communities and processes in a different way than we are used to. We should look for microbial traits such as substrate utilization or growth rate instead of only for taxonomy, understanding better the biochemistry of soil and find better ways how to address enzymatic processes or how to replace them with metabolomics approaches. Scaling down from the view of complex communities in large, artificially homogenised samples to individual isolates of microbes and from bulky soil core data to the metabolism of individual microhabitats is the key for understanding of the soil from the viewpoint of individual microbes. Contribution of understudied groups of soil biota: protists, nematodes, viruses (and who knows what else) to the complexity of soil processes needs to be built from a real scrap in the future. Undoubtedly, scaling up in space and time is also needed if we want to provide answers about ecosystem functioning valid for more than a few square meters at a single timepoint. The view that soil microbial ecology research is much more simple now with novel, high-throughput tools is an illusion. Soil will always be extremely complex and this will keep soil ecology a difficult but interesting and hopefully rewarding field of research.

FUNDING

This work was supported by the Czech Science Foundation projects No. 16-08916S and 18-25706S and by the Ministry of Education, Youth and Sports of the Czech Republic project No. LTT17022.

Conflict of interest. None declared.