One long debated question in community ecology is which processes determine an ecological community to assemble (Preston 1948). The current, well-established theories are primarily derived from research on animals and plants. Studies on microorganisms are scarce because of the assumption that microorganisms do not have distribution patterns due to their large numbers and small size (Baas-Becking 1934). However, using advanced sequencing technologies and statistical methods, microbial distribution patterns have been discovered in many natural environments including seawater and marine sediments (Liu et al. 2014b, 2015a; Lozupone and Knight 2007; Martiny et al. 2006). The spatial turnover of microbial communities always refects a distance-decay relationship and/or a taxa-area relationship, which are the two most well established patterns depicting increasing community dissimilarity with spatial distance (Nekola and White 1999) and increasing taxa richness with area size (Horner-Devine et al. 2004), respectively.
A framework has been established that considers the niche and neutral theories as potential mechanisms underpinning microbial biogeography (Dini-Andreote et al. 2015; Dumbrell et al. 2010; Leibold and McPeek 2006; Table 1). The long standing niche theory shows how biodiversity is structured by physiochemical (environmental heterogeneity) and biotic (inter taxa interaction) factors (Chase and Leibold 2003). In niche theory, every taxon is assumed to have unique and non-overlapping traits, which enable it to exert diferent responses/efects on an environment and to occupy diferent ecological niches (Leibold 1995). This heterogeneity in traits also eliminates inter taxa competition by resource partition and thus allows an infnite number of taxa to coexist in an environment, leading to high microbial diversity (Leibold and McPeek 2006). In this context, niche theory cannot explain the observed diferences in the abundance of diferent taxa, although microbial abundance under niche theory always follows a log-normal distribution pattern (Hubbell 2001). Later, Hubbell (2001) proposed a neutral ecological view on the assembly of taxa-rich communities that consisted of many rare taxa. Unlike niche theory, the neutral theory assumes that all individuals are ecologically equivalent, share the same way of life and have a similar response/efect on environments. Therefore, interactions among microbes, and between microbes and environments are ignored in the neutral theory (Hubbell 2001). Microbial diversity is controlled by randomly occurring events, such as ecological drift (change in the relative abundance of taxa in a location due to chance demographic fuctuations) and dispersal (movement of taxa across spaces) (Chave 2004; Hanson et al. 2012; Hubbell 2001). Because such ecological events happen diferently to each individual taxon, the relative abundance of diferent taxa can be partially explained by the neutral theory (Volkov et al. 2003).
The niche theory The neutral theory Time proposed 1959 (A general description of niche concept) 2001 Concept Coexistence by niche diferentiation Random events occurring on individual taxon Traits Functionally diferent Functionally equivalent Relative abundance Log-normal distribution Zero-sum multinomial distribution Processes Abiotic environmental selection and biotic interspecifc interactions Ecological drift and dispersal Feature of process Deterministic Stochastic
Table 1. Comparison of the niche and neutral theories
Microbial ecology studies have consistently considered abiotic selection as a niche-based (deterministic) process (also known as habitat flters or species sorting) and ecological drift and dispersal limitation as neutral-based (stochastic) processes (Fig. 1). Selections are mostly represented by environmental correlations. Drift is difcult to measure and it often interacts with restricted dispersal (dispersal limitation) to produce a distance-decay relationship (Hanson et al. 2012). Dispersal limitation is often represented by spatial distance. However, spatial distance can be a deterministic factor when it correlates to physiochemical factors and both population size (neutral) and traits (niche) can afect the dispersal ability of diferent taxa (Hanson et al. 2012). These make the relationships between selection, dispersal and stochasticity complex (Evans et al. 2017). To accurately evaluate stochasticity, neutral and null models have been increasingly implemented (Chase and Myers 2011; Jeraldo et al. 2012, Table 2). Nevertheless, these models cannot distinguish diferent aspects of the stochastic process. To solve this problem, Stegen et al.(2013, 2015) proposed a framework to quantify the relative roles of deterministic and stochastic processes. This framework integrates both the null model and phylogenetic information and assumes that phylogenetically close taxa tend to have similar ecological traits. It contains two steps with the frst to estimate the role of selection using phylogenetic dissimilarities and the second to diferentiate the role of dispersal and drift using the Bray–Curtis dissimilarities among microbial communities (Jia et al. 2018; Logares et al. 2018; Stegen et al. 2013, 2015; Zhou and Ning 2017).
Figure 1. The assembly of microbial communities is governed by a joint efect of deterministic (niche) and stochastic (neutral) processes. Subdivision of a microbial community (according to abundance, activity, function and occupancy) into subcommunities will facilitate an accurate view of microbial assembling processes
Community Ecosystem Spatial scale Stochasticity measurement Relative importance References Determinism Stochasticity Terrestrial ecosystem Arbuscular mycorrhizal fungi Soil < 50 m Species relative abundance and spatial distance √ – Dumbrell et al.(2010) Bacteria Desert Global Neutral model Heterotrophic community Photosynthetic community Caruso et al. (2011) Bacteria Freshwater lakes ~130 km in maximum Neutral community model – √ Roguet et al. (2015) Bacteria Soil spanning 105 years succession – Newly designed model Community when succession proceeded Community at initial stage Dini-Andreote et al.(2015) Bacteria and archaea Groundwater – Stegen's framework √ – Graham et al.(2017) Marine ecosystem Microbial eukaryotes Coastal water and sediment ~ 20 km in maximum Neutral community model and variation partitioning √ Chen et al. (2017) Bacteria Coastal water ~200 km Stegen's framework High eutrophication Low eutrophication Dai et al. (2017) Bacteria and protist Seawater 51–826 km Stegen's framework and variation partitioning Protist community(bottom water) Protist (surface and deep chlorophyll maximum waters) and bacterial (all three layers) communities Wu et al. (2018) Bacteria Coastal water ~1300 km in maximum Neutral model and variation partitioning – Both abundant and rare communities Mo et al. (2018) Archaea Coastal water ~200 km Stegen's framework and variation partitioning – √ Wang et al. (2019)
Table 2. Examples of studies examining the relative role of deterministic and stochastic processes in structuring microbial community
Deterministic and stochastic processes jointly govern the assembly of microbial communities (Chave 2004). However, their relative importance varies across diferent spatial and temporal scales (Table 2), depending on the strength of environmental gradients and the sensitivity of the microbes to environmental changes. If the extent of environmental variation is greater than the threshold a microbe can endure, dispersal will be prevented (Wang et al. 2013), leading to the predominance of determinism. Thus, the mechanisms underlying microbial community assembly would alter over a seasonal or longer term period with changes in the magnitude of environmental heterogeneity (Dini-Andreote et al. 2015; Langenheder et al. 2012). These conclusions are mostly based on investigations of terrestrial microbial communities. By comparison, there have been very few studies on the relative roles of deterministic and stochastic processes in the marine environment. However, there is a general perception that stochastic processes have a greater efect on the assembly of planktonic bacterial and archaeal communities than deterministic processes (Table 2). This can be explained by marine prokaryotes having evolved strong adaptation capabilities to environmental changes and by spatial connectivity and seawater movement homogenizing environmental conditions. Another explanation is that the environmental factors analyzed to represent deterministic processes may be not the most relevant ones afecting community variations. Further studies are needed to confrm such a hypothesis and to compare the assembling processes between diferent habitats, such as coastal water vs open ocean and water vs sediment.
Different types of marine organisms differ in their responses to deterministic and stochastic processes. Wu et al. (2018) reported that determinism had a stronger efect on planktonic protist communities than on bacterial communities, which may relate to diferences in their environmental sensitivity. Such diferent responses between bacteria and micro-eukaryotes have also been observed in soil (Powell et al. 2015a) and freshwater (Logares et al. 2018) habitats. Additionally, subcommunities that are divided by abundance, activity, functional trait or occupancy, can also undergo diferent ecological processes (Fig. 1). A microbial community is usually made up of a few abundant taxa and a long tail of rarer ones (Pedrós-Alió 2006, 2012). The rare taxa account for a great proportion of the microbial diversity and have been shown to assemble non-randomly and display similar distribution patterns to the abundant taxa (Galand et al. 2009; Gong et al. 2015; Liu et al. 2015b; Mo et al. 2018). Nevertheless, the abundant and rare taxa have both been observed to be diferently afected by stochastic and deterministic processes (Liu et al. 2015b; Mo et al. 2018). Mo et al. (2018) reported that the rare bacterioplankton in coastal seawater had a weaker response to environmental factors than abundant taxa; this may be due to the small population size of the rare taxa, making them more susceptible to ecological drift (Nemergut et al. 2013). By contrast, a survey of bacterial communities in freshwater lakes and reservoirs revealed a greater infuence of environmental changes on the rare than the abundant taxa (Liu et al. 2015b). These fndings suggest complicated microbial ecological responses across distinct ecosystems. In this context, further studies are needed to gain an insight into the assembling processes of abundant and rare taxa in different environments. The most urgent need is to propose a common defnition for rare taxa, facilitating a parallel comparison across studies (Jia et al. 2018). Subcommunities divided by functional traits, activity and occupancy receive less attention and have mainly been analyzed in terrestrial environments. For example, in deserts, the phototrophic community was mainly afected by stochastic processes, whereas the heterotrophic community displayed patterns mainly driven by environmental stresses (Caruso et al. 2011). The assembly of generalists and specialists in plateau lakes, however, was driven by stochastic and deterministic processes, respectively (Liao et al. 2016). While these studies provide novel and accurate information about the distribution patterns and assembling processes of microbial communities, there is an urgent need to investigate diferent subcommunities in the marine environment. It should be noticed that when evaluating the relative role of determinism and stochasticity, the estimated contribution from determinism is largely afected by the set of environmental factors measured, since they are not necessarily the most relevant parameters that provide the best explanatory power for the community variations.
In niche theory, the microbe-microbe interactions, although being ecologically important, are less understood compared to the microbe-environment relationships (Chase and Leibold 2003). Inclusion of interactions to explain microbial distribution patterns is a great challenge, largely due to the difculty in obtaining microbial co-cultures. An alternative way of elucidating microbial interactions is to apply correlation-based network analysis (Barberán et al. 2012; Layeghifard et al. 2017; Weiss et al. 2016), which is enhanced by the increase of community compositional data and the development of statistical tools. The most popular method used for constructing a correlation-based network is to calculate the Spearman's rank correlation coefcients between taxa (Barberán et al. 2012; Table 2). Other methods are also available, including SPIEC-EASI, CCLasso, REBACCA, CoNet, SparCC, WGCNA, Molecular Ecological Networks Analysis, Local Similarity Analysis, Maximal Information Coefcient, etc. These methods, however, vary in their sensitivity and precision (Layeghifard et al. 2017; Weiss et al. 2016).
Nodes and edges are fundamental components of a network, representing taxa and correlations, respectively. Edge thickness often denotes the degree of a correlation, with a thicker edge representing a higher correlation coefcient. On the basis of nodes and edges, a number of parameters can be calculated to represent the topological structure of a network, including degree, density, betweenness centrality, network diameter and clustering coefcient (Newman 2003). The degree of a node describes its connectivity to other nodes, with a higher value indicating a wider correlation. The betweenness centrality of a node describes the number of shortest paths between any two nodes going through it. The nodes with high degree and low betweenness centrality potentially represent the keystone taxa of a community (Berry and Widder 2014; Liang et al. 2016). The keystone taxa are the cornerstone and initial components for a community to assemble (Berry and Widder 2014) and have recently been defned as "highly connected taxa that individually or in a guild exert a considerable infuence on microbiome structure and functioning irrespective of their abundance across space and time" (Banerjee et al. 2018). A group of densely connected nodes with weak correlations to other nodes forms a module. Modular analysis can help to simplify the processes of identifying keystone taxa and/ or exploring the efect of environmental factors on microbemicrobe interactions.
Despite having the potential to infer mutualistic (positive) and antagonistic (negative) efects, the co-occurrence patterns illustrated by the network analysis can have diferent meanings, such as similar environmental preference and lifestyle, resource partitioning and nutrient cross-feeding that do not involve direct interactions. Network analyses have consistently revealed patterns dominated by positive cooccurrences in both terrestrial and marine bacterial communities (Barberán et al. 2012; Ju et al. 2014; Liu et al. 2014b; Ma et al. 2016; Milici et al. 2016; Zhang et al. 2014; Zhou et al. 2018). This suggests a ubiquitous non hostile process in bacterial community assembling irrespective of habitat. In fact, a network analysis of prokaryotic communities in global surface seawater samples (Tara oceans) showed that positive co-occurrences accounted for up to 90% of all correlations (Lima-Mendez et al. 2015; Milici et al. 2016). This may be explained by the similar environmental preference (possibly driven by some unknown/unmeasured environmental factors) and/or high resistance of marine prokaryotes to environmental stresses, which enable them to coexist in the same ecological niche. Indeed, the fnding is in line with the observation of a relatively weaker efect of determinism relative to stochasticity on the distribution of marine planktonic prokaryotes as stated above. It is also likely that auxotrophy in marine microbes contributes to the observed positive correlations, since several bacterial groups (such as the SAR11 clade of Alphaproteobacteria, Tripp et al. 2009) have been observed to gain ftness by obtaining a biomolecule from other groups. By contrast, a study of planktonic bacterial communities in the South China Sea showed that negative correlations dominated the co-occurrence patterns in active bacteria compared to positive correlations in total bacteria (Zhang et al. 2014). Resource competition may occur in diferent active bacterial groups. It is noticeable that microbial networks vary with space and time (Table 3). Signifcant changes in network topological structures have been observed in seawater at diferent depths (Chow et al. 2013) and between diferent seasons (Chafee et al. 2018). Additionally, Milici et al. (2016) reported that free-living bacterioplankton possessed highly interconnected networks compared to particle-attached communities. They postulated that the distinct nutrient-utilizing strategies of these two groups might be responsible for such a discrepancy. However, it is unexpected to fnd more between-taxa connections in the free-living community, since particle-attached microbes have a higher cell abundance and are physically closer to one another. This provides further evidence that the network-derived co-occurrence patterns are not always good proxies of true interactions between microbes.
Community Ecosystem Infuential factors Statistical methods References Network within a single system Bacteria and archaea Sediment – Spearman's rank correlations Liu et al. (2014b) Bacteria Sediment – Whittaker's index of associations Buttigieg and Ramette (2015) Bacteria Seawater – WGCNA Bryant et al. (2016) Bacteria Sediment – Spearman's rank correlations Qiao et al. (2018) Bacteria, archaea and microbial eukaryotes Seawater – Spearman's rank correlations Zhou et al. (2018) Network across diferent systems Spatially resolved network Bacteria Seawater Water depth Extended local similarity analysis Cram et al. (2015) Bacteria Seawater Particle size SparCC Milici et al. (2016) Bacteria and archaea Estuary Water and sediment Spearman's rank correlations Wei et al. (2016) Bacteria Seawater Water depth Spearman's rank correlations Cui et al. (2019) Temporally resolved network Bacteria and archaea Seawater (deep chlorophyll maximum) Month Local similarity analysis Beman et al. (2011) Bacteria Seawater Month Extended local similarity analysis Cram et al. (2015) Bacteria, myoviruses and phytoplankton Seawater Daily to weekly Local similarity analysis Needham et al. (2017) Protist Seawater Daily to weekly Extended local similarity analysis Berdjeb et al. (2018) Bacteria and archaea Seawater Season SparCC Chafee et al. (2018)
Table 3. Examples of network analysis conducted in the marine ecosystem
Several studies have attempted to use networks to infer potential functional couplings between microbes. For example, Thaumarchaeota Marine Group I (MG-I), the most abundant archaeal clade in the marine environment, capable of ammonia oxidization, has been found to co-occur with Nitrospina (Reji et al. 2019) and/or with Nitrospira when Nitrospina is absent or in low abundance (Wang et al. 2019), both of which are nitrite oxidizers. Their co-occurrence in seawater is supported by substrate feeding (nitrite produced by MG-I is the substrate of Nitrospina/Nitrospira) and facilitates the complete nitrifcation process. Previous eforts to explore co-occurrence patterns between functional bacteria in marine sediments have demonstrated signifcant correlations between sulfate-reducing bacteria and sulfuroxidizing bacteria, and between sulfate-reducing bacteria and nitrite-oxidizing bacteria (Liu et al. 2014b). Elucidation of co-occurrence patterns with functional gene abundance derived from GeoChip and metagenome may facilitate a more direct inference. However, the obtained co-occurrence patterns should be treated with caution when used to infer functional couplings, since they are not necessarily refecting real interactions.
Classically, a network describes co-occurrence patterns between taxa. However, environmental variables can also be included to explore microbe-environment relationships. Additionally, considering the natural complexity of inter taxa relationships in an ecosystem, pairwise microbe-microbe correlations, derived from most current network analyses, need to be expanded to a higher order, such as three- or fourway correlations. The high-order microbial co-occurrence patterns may involve possible disruption or enhancement of another taxon to a pairwise relationship (Bairey et al. 2016). Such high-order co-occurrence patterns can also be unraveled by analyzing compositional data, as long as new and proper statistical tools are developed (Bairey et al. 2016). Although co-occurrence patterns are not appropriate to imply accurate microbial interactions, their spatiotemporal dynamics hold the potential to afect the assembling processes and ecological roles of microbial communities.