Keystone Taxa As Drivers of Microbiome Structure and Functioning

PERSpECTIVES

influence in the community33. On the basis OPINION of the available information, in addition to sea stars, other examples of keystone taxa Keystone taxa as drivers of include the Canadian beaver and African elephant in the animal kingdom and leguminous Trifolium in the plant kingdom. microbiome structure and In microbial communities, examples of such taxa are now available from a diverse functioning range of environments24–26,34–61, and their reports are continuously increasing (Box 1), Samiran Banerjee, Klaus Schlaeppi and Marcel G. A. van der Heijden including Porphyromonas gingivalis and Bacteroides thetaiotaomicron in the human Abstract | Microorganisms have a pivotal role in the functioning of ecosystems. microbiome56,62–64. B. thetaiotaomicron, an Recent studies have shown that microbial communities harbour keystone taxa, anaerobic symbiont found in the human which drive community composition and function irrespective of their abundance. intestine, is considered a keystone taxon In this Opinion article, we propose a definition of keystone taxa in microbial based on empirical evidence56,58. Owing ecology and summarize over 200 microbial keystone taxa that have been identified to the complexity of microbial communities, the importance of connectedness and the in soil, plant and marine ecosystems, as well as in the human microbiome. We rapid turnover in both time and space, explore the importance of keystone taxa and keystone guilds for microbiome the definition of keystone taxa for structure and functioning and discuss the factors that determine their distribution microbiology needs to be adapted and activities. from the original concepts proposed in ecology30–32. In this Opinion article, we propose the following definition: microbial The role of microbial communities in demonstrated that the removal of sea stars keystone taxa are highly connected taxa ecosystem functioning is unequivocal1,2, (Pisaster ochraceus, which is a common that individually or in a guild exert a with microorganisms being key drivers predator of mussels) had a dramatic impact considerable influence on microbiome of many ecosystem processes, including on the shoreline ecosystem community structure and functioning irrespective of soil nutrient cycling, plant growth, marine and local biodiversity at Makah Bay, their abundance across space and time. biogeochemical processes and maintenance Washington, USA28. Since the term was These taxa have a unique and crucial role in of human health3–6. In recent years, first coined, the definition of keystone taxa microbial communities, and their removal microbial network analysis has been used has followed different lines of thought29–31. can cause a dramatic shift in microbiome to visualize co-occurrence among members The definition proposed by Paine in structure and functioning. in communities3,7–10. Microbial network 1969 mainly suggests that keystone taxa We briefly discuss how microbial analysis enables testing of ecological are important for community structure network analysis can be used to identify theories, the assessment of which was once and integrity, and their influence is non- keystone taxa and focus on recent evidence postulated to be a major impediment in redundant32. In 1996, a study defined of keystone taxa based on computational microbial ecology11,12. The concept of co- keystone taxa by introducing the concept inference and empirical evidence. We also occurrence and network thinking in ecology of ‘community importance’, which was discuss challenges in identifying keystone was proposed in 2005 (ref.13), and since then, calculated from proportional biomass taxa, including the characterization and microbial ecologists have shown particular and traits31. Subsequently, in 2012, other manipulation of such taxa, and explore their interest in network analysis7,14–19, resulting authors30 presented the evolution of the influence on microbiome functioning as in a large body of studies demonstrating term keystone taxa in ecology and how well as the factors that may determine their microbial co-occurrence patterns in a its overuse and misuse (for example, distribution and efficacy. diverse range of soil7, plant20 and marine21 keystone mutualist, keystone modifier ecosystems, as well as in the human and reverse keystone) have resulted in Microbial networks and keystone taxa microbiome22,23 (Box 1). Reports are also considerable confusion about the actual With the advent of next-generation available from the Antarctic ecosystem24 meaning (readers are referred to their sequencing, millions of sequences are and Arctic ecosystem25,26. In addition to critical appraisal for further information now available from various environments. co-occurrence patterns, microbial networks on keystone taxa in ecology). Thus, there Network analysis can help disentangle can be used to statistically identify is no uniformly accepted operational microbial co-abundance and provide keystone taxa27. definition of keystone taxa in ecology, comprehensive insight into the microbial The tenet of keystone taxa was originally especially in microbial ecology. Keystone community structure and assembly proposed by ecologist Robert T. Paine taxa have also been frequently referred to as patterns3,65. Several algorithms are available in 1966. In a classic experiment, he ‘ecosystem engineers’ owing to their large to construct microbial networks, and

Box 1 | Microbial co-occurrence and network analysis keystone taxa belong to the Pezizomycotina subdivision38. Keystone taxa that were the number of papers (based on the web of science database) reporting microbial network not numerically dominant in the analyses and keystone taxa is increasing exponentially (see the figure; data are based on the web of communities have also been identified in science database, time span 2004–2016, keywords ‘microbial network analysis’ and ‘microbial the Arctic ecosystem25,26,44,46 and Antarctic keystone’). a wide range of methods and algorithms are available to construct microbial networks. 24 starting with basic Pearson or spearman rank correlation-based approaches7,14, microbial networks ecosystem . Similar reports of numerically quickly evolved to incorporate more robust methods. For example, the maximal information inconspicuous keystone taxa are also coefficient (MiC) relies on equitability and generality of relationships15 and can yield a variety of available for microbial communities in linear and nonlinear associations among microorganisms that can be interpreted similarly to the contaminated soils47,48, roots71 and aquatic coefficient of determination (denoted as r2). Local similarity analysis (Lsa) that detects change in systems24,55,72. Interestingly, our literature abundance of operational taxonomic units (Otus) over time in an environment is particularly useful review revealed that various members of for analysing microbial temporal variability the Rhizobiales and Burkholderiales orders data16, whereas sparse correlation for were consistently identified as keystone 0GVYQTMCPCN[UGU compositional data (sparCC) is especially suited taxa in different studies and across different for compositionally diverse microbial data17. ecosystems (Table 1; see Supplementary Moreover, the ensemble approach (CoNet) can use multiple measures (similarity, correlation Table 1). Rhizobiales comprises not only and mutual information) with the generalized nitrogen-fixing bacteria, such as Rhizobium boosted linear model to generate spp. and Bradyrhizobium spp., but also comprehensive networks18. a recent study19 members of the genus Methylobacterium, U reported that Lsa, MiC and sparCC are well which is known to be endosymbiotic suited for both count and compositional data and abundant in the phyllosphere73. and are less sensitive to how data are By contrast, Burkholderiales includes distributed, whereas the CoNet ensemble important genera such as Bordetella, /KETQDKCNMG[UVQPGVCZC approach performs better for data with high

0WODGTQHRCRGT Ralstonia and Oxalobacter, which are well- scatteredness or sparsity. However, it was also known pathogens, as well as Burkholderia, noted that different approaches may yield different results and significance levels for the which is one of the most versatile and same data set, and although the scores obtained diverse terrestrial microbial groups. The from thousands of pairwise correlations are computational identification does not typically corrected for type i error using a mean that all members of the Rhizobiales Bonferroni correction or the false discovery rate, and Burholderiales can be considered extremely rare Otus or Otus with a large keystone taxa (for example, many taxa number of zeros should be avoided for network in those orders are subordinate taxa in 2WDNKECVKQP[GCT construction. microbial communities and have no major influence on community composition or functioning). Computational inference these algorithms have been reviewed these scores have been used to find putative of Rhizobiales and Burkholderiales as previously19,65,66; thus, for brevity, we only keystone taxa in microbial networks in keystone taxa can also be due to their present a brief overview (Box 1). Perhaps recent studies12,41,42. The importance of sheer abundance in various environments. one of the most useful features of network a quantifiable threshold for consistent Nonetheless, the likelihood of finding a analysis is that ‘hubs’ (also termed keystone identification and validation of keystone keystone taxon within these two orders is operational taxonomic units (OTUs)), taxa has been highlighted30, and we high, and future studies are now needed to which are taxa that are highly associated recommend that the combined score of high evaluate the role of putative keystone taxa in in a microbiome (Fig. 1), can be identified. mean degree, high closeness centrality and microbial functions. Unlike random networks with a Poisson low betweenness centrality27 should be used distribution, scale-free or small-world as a threshold for defining keystone taxa in Empirical evidence. Human microbiome networks with a power-law distribution microbial communities. studies have provided most of the empirical comprise such hubs (also referred to as evidence of keystone taxa, linking keystone highly connected nodes; reviewed in Recent evidence of keystone taxa taxa to a range of processes, including refs67,68). These hubs have been proposed Computational inference. Numerous inflammation, colon and gastric cancer, as keystone taxa, as their removal has been studies have used network-based scores to starch degradation and stabilization of the computationally shown to cause a drastic identify putative keystone taxa in various human-associated microbiota22,23,56–60,64,74–76 shift in the composition and functioning of environments (Table 1; see Supplementary (Supplementary Table 1). Perhaps one a microbiome69,70. Thus, network analysis Table 1). Hubs in microbial networks of the most prominent keystone taxa can be a powerful tool for inferring keystone were identified in grassland soils, and the in humans is Bacteroides fragilis, which taxa from microbial communities. Although Pampa and Cerrado biomes in Brazil were spurred the alpha-bug or keystone pathogen high betweenness centrality was previously shown to harbour different keystone taxa hypothesis59,74. Other examples of keystone used to identify keystone taxa statistically in mostly belonging to the Actinobacteria taxa in humans include P. gingivalis64, several studies16,39,40, it was recently shown and Proteobacteria phyla35. A network B. thetaiotaomicron56, Ruminococcus bromii60, that high mean degree, high closeness analysis at the continental scale showed Methanobrevibacter smithii74 and Helicobacter centrality and low betweenness centrality that bacterial keystone taxa are members pylori57. Both observational and manipulative can be collectively used to identify keystone of the Alphaproteobacteria class and studies of the gut microbiome show that taxa with 85% accuracy27. Subsequently, Actinobacteria phylum, and fungal these taxa can exert considerable control

C 0GVYQTMYKVJQWVMG[UVQPGVCZCQTOQFWNG 0GVYQTMYKVJMG[UVQPGVCZCDWVPQOQFWNG 0GVYQTMYKVJMG[UVQPGVCZCCPFOQFWNGU

&[UDKQUKUKPQTCNOKETQDKQOG 2QTRJ[TQOQPCUIKPIKXCNKU

%QPUKFGTCDNGCNVGTCVKQPKPIWV +ORTQXGFRNCPVGXGPPGUU .KPMGFVQUQKNQTICPKEOCVVGT OKETQDKQOGUVTWEVWTG CPFRTQFWEVKXKV[ FGEQORQUKVKQP *GNKEQDCEVGTR[NQTK 4JK\QDKWOURR )GOOCVKOQPCUCPF#EKFQDCEVGTKC $CEVGTKQFGVGUVJGVCKQVCOKETQP 4WOKPQEQEEWUDTQOKK $CEVGTQKFGUHTCIKNKU Fig. 1 | Keystone taxa in the microbiome. a | Modularity and keystone taxa keystones may cause a dramatic shift in the composition70. b | Empirical evi- in microbial networks. Nodes (small red dots) represent operational taxo- dence of keystone taxa in the human (left panel), plant (middle panel) and nomic units (OTUs), and solid lines represent edges, that is, relationships soil (right panel) microbiomes. Smaller dots and ovals represent general among nodes. A network consisting of many taxa (nodes), without any members, whereas larger dots represent keystone taxa in the community. In highly interacting keystone taxa (left panel), is similar to a random network the oral microbiome, Porphyromonas gingivalis causes inflammatory tissue that has a Poisson distribution of edges per node, that is, most nodes have destruction and initiates imbalance or dysbiosis of the community that similar number of edges and no highly connected nodes. A microbial net- favours further growth of this keystone taxon74. Similarly , in the human gut work without any modules but with keystone taxa (large black dots) is shown microbiome, keystone taxa, such as Bacteroides thetaiotaomicron56, in the middle panel. This is a scale-free network that has a power-law distri- Bacteroides fragilis59, Helicobacter pylori57 and Ruminococcus bromii60, exert bution of edges, that is, only two nodes are highly connected and holding considerable control on microbiome structure and functioning. Nitrogen- the network together. A module is a cluster of highly interconnected nodes. fixing rhizobia have been proposed as keystone taxa as their abundance can A network with two highly connected keystone taxa (large black nodes) that improve plant productivity and community evenness77. In the soil micro are positioned in two distinct modules is shown in the right panel. These biome, bacterial and fungal keystone taxa identified using network scores keystone taxa are holding the modules together. Thus, removal of such have been found to be linked to organic matter decomposition39,41.

on the composition and functioning of the shown to be mediated via microbial keystone linked to organic matter decomposition in oral and gut microbiome. Examples are also taxa20. This not only supports the relevance of an agricultural soil39. These taxa were also available from plant and soil microbiomes, keystone taxa but also provides evidence identified as keystones for soil organic matter where keystone taxa have been identified of their importance for plant microbiome transformation in another study41, indicating through network-based scores and linked functioning. Nitrogen-fixing rhizobia the importance of similar keystone taxa for to microbiome functioning and ecosystem have been proposed as keystone taxa, and specific habitats and processes. Moreover, we processes. The effect of abiotic factors (for their abundance has been shown to greatly predict that important plant symbionts, such example, sampling time and temperature) improve plant productivity and community as mycorrhizal fungi, function as keystone and host genotypes on phyllosphere microbial evenness77. Fungal and bacterial keystone taxa in view of their role as ecosystem communities in Arabidopsis thaliana were taxa were recently identified that were engineers and their impact on microbial

Table 1 | Summary of studies reporting keystone taxa in different ecosystems communities, plant diversity and ecosystem functioning78,79. A recent study found that Ecosystem or habitat Keystone taxaa refs low abundance keystone taxa that are highly Computational inference connected in the microbiome can explain 34–36 Grasslands • Burkholderiales microbiome compositional turnover better • Sphingobacteriales 80 • Clostridiales than all taxa combined . Indeed, such • Actinomycetales encouraging reports highlight the relevance • Acidobacteria GP4 of keystone taxa for microbiome composition Forest or woodlands • Actinomycetales 8,35,37,38,61 and functioning. • Acidobacteria GP4 • Rhizobiales Challenges in identifying keystone taxa • Burkholderiales Correlation does not imply causation. • Clostridiales • Sphingobacteriales Keystone taxa identified using network- • Rhodobacterales based scores have been linked to ecological • Verrucomicrobia processes in many studies (Supplementary Agricultural lands • Gemmatimonas 35,40,42,43 Table 1), indicating that this is a suitable • Acidobacteria GP17 approach. However, network-based scores • Xanthomonadales • Rhizobiales need to be complemented with experimental • Burkholderiales evidence showing the impact of the keystone • Solirubrobacterales taxa on microbiome composition and • Verrucomicrobia function. The detection of keystone taxa Arctic and Antarctic ecosystems • Rhizobiales 25,26,44,46 using network-based scores alone can be • Burkholderiales biased by habitat filtering, and networks • Actinobacteria can display positive associations between • Alphaproteobacteria non-interacting microbial members in Contaminated soil • Rhizobiales 47,48 • Nitrospira environmental samples. Moreover, network • Pseudomonadales scores and co-occurrence patterns are • Actinobacteria ultimately based on correlations, and Plant-associated microbiota • Acidobacteria GP1, GP3 and GP6 40,49,50 they must be interpreted with caution, • Rhizobiales as correlation does not mean causation. • Burkholderiales Statistical analyses such as structural • Pseudomonadales • Bacteroidetes equation modelling (SEM) can be used • Frankiales to move beyond correlation analysis and Aquatic ecosystems • Pelagibacter 24,51–55,72 explore causal relationships among keystone • Oceanospirillales taxa and microbiome composition or • Flavobacteriaceae function. SEM is an advanced multivariate • Nitrospira statistical approach that identifies such • Rhodobacteradaceae • Alteromonadaceae causal relationships and generates strong • Chromatium and distinct links between theoretical and • Rhizobiales experimental ideas81. The strength of SEM • Burkholderiales lies in the fact that it is theory oriented • Chlorobium and not null hypothesis based, and thus, it • Verrucomicrobia • Chloracidobacterium provides a framework to interpret complex • Chloroflexi networks involving numerous response • Candidatus OP3 and predictor variables. Upon assessing the Empirical evidence univariate and multivariate normality, an Agricultural landsb • Gemmatimonas 39,41 initial model is generated on the basis of • Acidobacteria existing knowledge, site information and Phyllosphere • Albugo 20 background data82,83. Subsequently, a χ 2 test is • Dioszegia conducted to assess whether the covariance Human oral microbiome Porphyromonas gingivalis 64,74 structure indicated by the model adequately Human gut microbiome • Helicobacter pylori 22,23,56–60,76 fits the actual covariance structure of the • Methanobrevibacter smithii data, with a nonsignificant χ 2 test result • Actinobacteria suggesting sufficient model fit. Importantly, • Bacteroides fragilis SEM requires a minimum sample size of 50 • Bacteroides stercoris 81 • Bacteroides thetaiotaomicron (ref. ). Moreover, determining the relative • Ruminococcus bromii influence of keystone taxa can also be a • Klebsiella pneumoniae challenge84. A recent study used sparse linear • Proteus mirabilis regression with bootstrap aggregation in aMembers of Rhizobiales and Burkholderiales are consistently present across ecosystem types, except for the a discrete-time Lotka-Volterra model to human microbiome. Keystone taxa across individual studies and links between keystone taxa and specific microbial ecosystem processes as reported in relevant studies are shown in Supplementary Table 1. bKeystone identify B. fragilis and Bacteroides stercoris taxa were initially identified using network-based scores and linked to organic matter decomposition. as keystone taxa with disproportionate

influence on the gut microbiome structure22. Influence on the microbiome and the release of nutrient-rich exudates, Using a novel approach called Learning Influence, irrespective of abundance, initiating dysbiosis of the oral microbiota Interactions from MIcrobial Time Series distinguishes keystone taxa from dominant and favouring further growth of (LIMITS) on metagenomic data, it was taxa. A dominant species often affects P. gingivalis74,94. Similarly, in the case statistically shown that the moderately ecosystem functioning or a specific process of inflammatory bowel disease or Crohn’s abundant B. fragilis and B. stercosis can exert exclusively by virtue of sheer abundance disease, dysbiosis results in reduced diversity significant influence on the microbiome, (Fig. 2a), whereas keystone taxa might of major phyla such as Bacteroidetes and and any perturbations applied to these taxa exert their influence on microbiome Firmicutes and increased abundance of have a large impact on microbial community functioning irrespective of abundance. Enterobacteriaceae93,95. structure. This is encouraging because the The importance of keystone taxa may also In the plant microbiome, certain algorithm identified B. fragilis as a keystone be related to the broadness of a process, strains of Pseudomonas fluorescens taxon, thus agreeing with existing that is, a process involving many steps as produce a secondary metabolite empirical data74. well as functionally and taxonomically (2,4-diacetylphloroglucinol) that suppresses diverse microbial groups1,91. For example, Gaeumannomyces graminis var. tritici, Characterization and manipulation. dominant taxa with large biomass or major which causes the take-all disease in Although experimental manipulation energy transformations might influence wheat96. Alternatively, keystone taxa might (for example, removing a putative keystone broad processes, such as denitrification or produce bacteriocins to selectively alter taxon to assess the impact) is the popular organic matter decomposition. By contrast, microbiota composition. For example, choice among plant and animal ecologists, the influence of rare keystone taxa might bacteriocin production by Enterococcus one of the fundamental challenges that be more pronounced if a process is narrow, faecalis can induce niche competition in the microbiologists are confronted with is consisting of a single step (for example, gastrointestinal tract to change microbiota the characterization and manipulation nitrogen fixation or ammonia oxidation) composition78. Keystone taxa might also of such taxa. Manipulating growth or and being carried out by a small group engage in synergistic relationships and co-culturing microorganisms on nutrient of specialized microorganisms1,91. We change the abundance of their partners, media or Petri dishes or in microcosms postulate that the influence of rare keystone and this could have an effect on community can be challenging owing to individual taxa on an ecosystem process is inversely structure and performance. Some physiological requirements. In past years, proportional to the broadness of the process. members of the Burkholderia genus can novel approaches have been developed However, it should be noted that some function as an endosymbiont in arbuscular to overcome the uncultivability issue. keystone taxa, such as B. thetaiotaomicron mycorrhizal fungi to change the abundance For example, the isolation chip, which in the human intestine, can be numerically and community characteristics of this is composed of numerous diffusion dominant, and thus, the distinction important fungi79, which subsequently chambers, enables in situ cultivation between dominant taxa and less abundant may alter plant community richness and of previously uncultivated microbial keystone taxa is not always clear. Whether productivity97. Although not considered species85. Similarly, a microbial trap has numerically inconspicuous keystone taxa yet, we hypothesize that mycorrhizal fungi been developed to capture and culture are more influential on narrow processes is a function as keystone taxa because these filamentous Actinobacteria under in situ hypothesis that needs further investigation. plant symbionts have a major impact on soil conditions86. Moreover, on-chip microbial Keystone taxa might use a range microbial communities, plant diversity and culture coupled with surface plasmon of strategies to exert an influence on a ecosystem functioning79. Thus, keystone resonance enables the in situ detection microbiome. For example, they might taxa can use different strategies to shape the of novel and rare microorganisms87. function via intermediate or effector microbiota in their favour, but the selection Droplet-based microfluidic technology groups, whose abundance can be selectively of a particular strategy would depend on also offers the opportunity to mimic modulated to regulate community the microenvironment. We speculate that natural conditions and co-cultivate structure and functioning23,74. Such many such strategies are aimed to gain synergistic microbial communities88, selective modulation might include direct benefits, such as replacing indigenous and the microbiome-on-a- chip approach promotion (commensalism) or suppression microflora (in case of E. faecalis), gaining enables the study of microbial networks (amensalism) of effector groups by secreting competitive advantage in the community and their associations with host metabolites, antibiotics or toxins. In humans, (in case of P. gingivalis) or promoting further plants89. Future studies may include P. gingivalis affects the community by causing growth (in case of B. thetaiotaomicron). such promising approaches to isolate dysbiosis, which results in inflammation However, it is possible that metabolites and characterize keystone taxa from and periodontitis64. Here, effector groups or by-products from keystone taxa may various environments and explore their are accessories used by keystone taxa to alter influence members of the microbiome with functioning. Removal of keystone taxa microbiome composition and manipulate indirect or even no benefits to the keystones. may lead to an alternative stable state their hosts92, and dysbiosis is the imbalance (sensu90) of the microbial network, which in the composition of the microbiome93. Putative drivers of keystone taxa results in dysfunction or even renewed In the case of chronic periodontitis, The presence of keystone taxa in a functioning if the removed keystone had P. gingivalis functions as the keystone taxon, microbiome does not necessarily guarantee a negative impact. Future studies may also whereas Streptococcus gordonii functions their influence because a number of factors enable the experimental manipulation as the accessory94. P. gingivalis impairs host may still determine their distribution and of microbial network structure in defence and causes overgrowth of the oral efficacy (Fig. 2b). For example, spatiotemporal synthetic communities to assess whether commensal bacterium S. gordonii. The heterogeneity can be a major driver of the the removal of keystone taxa disrupts co-adhesion of this keystone–accessory abundance and distribution of keystone microbiome functioning. pair causes inflammatory tissue destruction taxa29,31,84. This is particularly true for soil,

C &QOKPCPVVCZC +ORCEVFWGVQ +ORCEVQPCRTQEGUU E JKIJCDWPFCPEG KTTGURGEVKXGQHDTQCFPGUU 0CTTQY RTQEGUU /KETQDKQOG HWPEVKQPKPI

-G[UVQPGVCZC -G[UVQPGVCZC 5GNGEVKXG +ORCEVKPXGTUGN[ OQFWNCVKQP RTQRQTVKQPCNVQDTQCFPGUU

/KETQDKQOG HWPEVKQPKPI $TQCF RTQEGUU

D 5RCVKQVGORQTCN

JGVGTQIGPGKV[ QCF 2QUKVKQPKP $T OKETQDKQOG QEGUU +PFKXKFWCNXGTUWU MG[UVQPGIWKNF

(WPEVKQPCN QDKCNRT TGFWPFCPE[ /KET

%QPVGZV QY FGRGPFGPE[ 0C TT *[UVGTGUKU .QY *KIJ -G[UVQPGIWKNF +PȱWGPEGQHMG[UVQPGVCZC Fig. 2 | Keystone taxa in microbial communities and the factors influ- Spatiotemporal heterogeneity can drive the abundance and distribution of encing their functioning in an environment. a | The dominant taxa (light keystone taxa in any environment, especially in soil. The impact of a key- orange) affect microbiome functioning exclusively by virtue of sheer abun- stone taxon may be higher if it belongs to the core microbiome that is con- dance, whereas keystone taxa (green) exert their influence irrespective of sistently present in an environment regardless of changes in environmental their abundance. As the impact of dominant species on a process is primarily conditions. Keystone taxa may function alone, or a group of such taxa with due to greater abundance, the broadness of that process is less important. similar functioning may form a keystone guild and alter the structure and Here, broadness implies that a particular process consists of many steps and dynamics of the ecosystem that they thrive in. In the microbial world, it is involves diverse microbial groups. By contrast, keystone taxa exert their possible that keystone taxa may be functionally redundant or that they are influence by selectively modulating accessory microorganisms, and thus, only relevant in a particular context. Hysteresis suggests a time lag between they might have a greater influence on narrow processes (the processes that the functioning of a keystone and its detectable outcome in the micro consist of a single step or a few steps and involve a select group of micro biome. c | Hypothetical diagram showing various modes of functioning of organisms). The accessory microorganisms whose abundance is selectively keystone taxa in an environment. Individually , keystone taxa (teal dot) might promoted by keystone taxa are shown in blue, whereas other community have greater influence on a narrow process (for example, biological nitrogen members are shown in dark orange and purple. b | Environmental and eco- fixation performed by highly specific microorganisms). A keystone guild logical factors that may determine the distribution and performance of comprising multiple keystone taxa (yellow , black and purple dots) within a keystone taxa in an environment. The influence of keystone taxa on a micro- community might also be able to influence a broad process, such as organic bial process is inversely proportional to the broadness of that process. matter decomposition and denitrification. which is one of the most heterogeneous the first evidence of a core gut microbiome rat, which can be considered a keystone and multifaceted environments. Similarly, in obese and lean twins99. A recent study guild in the Chihuahuan desert, USA, and seasonal variability determines the reported an evolutionarily conserved core has a strong impact on local biodiversity structural and compositional properties microbiome in plant roots, and, intriguingly, and biogeochemical processes101. In the of microbiomes in an environment, and as some of the well-known keystone taxa, microbial world, keystone guilds may arise such, a keystone might be present only in a such as Rhizobium, Bradyrhizobium and on the basis of a number of factors, including, specific season or time period. Burkholderia, are also part of the core root for example, complementary resource The occurrence and functioning of a microbiome100. The contribution of keystone acquiring strategies, resource sharing, keystone will also depend on its position taxa will be higher if they are part of the core niche partitioning and spatiotemporal in the microbiome. Recently, the tenet of microbiome and consistently present in an coherence62,84. Whereas numerically core microbiomes and holobionts has been environment, highlighting the importance of inconspicuous keystone taxa might have proposed for humans6,98 and plants92, and such taxa for microbiome functioning23. a greater influence on narrow processes, a readers are referred to references6 and92 for Microbiomes can also harbour keystone keystone guild consisting of diverse keystone the taxonomic and functional definitions guilds (that is, groups of keystone taxa with taxa within a community might also influence of a core microbiome. Keystone taxa might similar functioning)31 (Fig. 2c). Examples of a broad process. For example, certain be part of the core microbiome that is such guilds that can alter the structure and keystone guilds of co-occurring denitrifiers consistently present in an environment dynamics of ecosystems are common in the can have an important role in denitrification, regardless of changes in environmental animal world101. Perhaps the most famous a broad process that involves heterogeneous conditions92,98. A seminal paper presented example is the three species of kangaroo groups of microorganisms102. We expect that

/KVKICVG01 GOKUUKQPU /CPKRWNCVG MG[UVQPGVCZC /CPKRWNCVG KPVJGJWOCP MG[UVQPGVCZC OKETQDKQOG KPFGPKVTKȰGT IWKNFU +ORTQXG *CTPGUUMG[UVQPG %QOOWPKV[ /KETQDKQOG JWOCPJGCNVJ VCZCKP51/ RTQȰNKPI OQFGNNKPI FGEQORQUKVKQP

'PJCPEGECTDQP -G[UVQPG UVCDKNKV[ VCZC

0GVYQTM 5+2CPF CPCN[UKU OGVCQOKEU *CTPGUUMG[UVQPG *CTPGUUMG[UVQPG VCZCKPOCTKPG VCZCKPRNCPV OKETQDKQOG OKETQDKQOG

2TGFKEV +ORTQXGRNCPV EQOOWPKV[UJKHV RGTHQTOCPEG /KETQDKQOG %WNVWTKPI CPFFKURGTUCN QPCEJKR

Fig. 3 | characterizing and harnessing keystone taxa. Hypothetical greenhouse gas, can be mitigated by manipulating denitrifier guilds. diagram illustrating the tools (hexagons) for linking keystone taxa to eco- Denitrification is a major contributor to nitrous oxide emissions. Similarly , system functioning and the research areas (circles) where keystone taxa can keystone taxa that are linked to soil organic matter (SOM) decomposition be used. Although network analysis can be used to statistically identify can be manipulated to enhance carbon sequestration in soil. Harnessing keystone taxa in microbial networks, it is important to link such taxa to keystone taxa in plant microbiomes can be valuable to enhance plant pro- ecosystem processes. With the advent of newer tools, such as chip or ductivity in agricultural systems or to alter performance of invasive plants. culture-based methods, keystone taxa can be isolated from environments Most of the empirical evidence on keystone taxa emerged from the human and cultured or co-cultured. Functional profiling of such taxa can be microbiome studies where keystone taxa such as Porphyromonas gingivalis, performed using RNA-stable isotope probing (SIP) coupled with metatran- Bacteroides fragilis, Bacteroides thetaiotaomicron and Ruminococcus bromii scriptomics or metaproteomics. Upon functional profiling, the relative have been identified. Targeted manipulation of these pathogens can facil- importance can be estimated through microbiome modelling. Such models itate medical interventions and improve human health. In aquatic systems, involving causal relationships can be used to reveal the contribution of key- it has recently been shown that keystone taxa can explain microbiome stone taxa to ecosystem processes. There are several areas where keystone compositional turnover better than all taxa combined80. Such keystone taxa taxa or guilds have been identified and thus can be harnessed for improved can be harnessed to predict shifts in the community or to manipulate ecosystem services. For example, emissions of nitrous oxide (N2O), a potent microbiome functioning. examples of such keystone guilds in microbial of keystone taxa and their influence on methanotrophy and sulfate reduction communities will continue to be identified microbiome functioning. With rapid (reviewed in refs103,104). For example, in the future. Indeed, such guilds may be microbial turnover, identifying such lags can Desulfosporosinus spp., which only represent particularly powerful if they belong to the be a daunting task. The above discussion 0.06% of the total community, have a pivotal core microbiome. of potential drivers of keystone taxa is not role in sulfate reduction and carbon flow Keystone taxa or members of keystone exhaustive, and there may be other factors in peatland soils105. The rare biosphere was guilds might be functionally redundant, or influencing these taxa in the microbial world. also shown to be important in the human their effect might be context dependent. microbiome and even in depauperate Such context dependency or conditionality The rare species concept ecosystems104. Evidence of such low abundant may be more common in environments Keystone taxa underline the importance taxa with an overproportional influence with turbulence or high spatiotemporal of numerically inconspicuous taxa for obviously raises the possibility that members variability31. Thus, keystone taxa may not microbiome functioning, which is also of rare biosphere can also be keystone taxa. always be present in an environment or may congruent with the rare taxa concept. not have the same impact on the community Indeed, the fundamental premise of Outlook under changing conditions. A taxon should keystone taxa and rare taxa is the same: Beyond the dominant taxa with large biomass only be considered a keystone in the context the abundance of a species is not the best and major energy transformations, keystone or condition under which it has a large determinant of its contribution to the taxa can orchestrate microbial communities influence. A plausible challenge for assessing community103. The importance of rare to perform ecosystem processes. This keystone taxa is also the fact that there microorganisms has been documented for Opinion article highlights the relevance might be a hysteresis effect, that is, a time many biogeochemical processes, including of keystone taxa as drivers of microbiome lag between the change in the abundance nitrification, denitrification, methanogenesis, structure and functioning. Owing to current

tautonyms and misconceptions surrounding Another intriguing question is whether 1. Fierer, N. Embracing the unknown: disentangling the complexities of the soil microbiome. Nat. Rev. keystone taxa, we proposed a definition keystone taxa in microbial communities Microbiol. 15, 579–590 (2017). of keystone taxa in microbiology. We also follow similar ecological principles (for 2. Bardgett, R. D. & van der Putten, W. H. Belowground biodiversity and ecosystem functioning. Nature 515, presented a summary of computational example, drift, dispersal, diversification and 505–511 (2014). inference and empirical evidence of over environmental selection; sensu12) as keystone 3. Fuhrman, J. A. Microbial community structure and its functional implications. Nature 459, 193–199 200 keystone taxa reported for soil, plant taxa in plant or animal kingdoms. (2009). and marine ecosystems and the human Linking community structure to function 4. van der Heijden, M. G. A., Bardgett, R. D. & Van Straalen, N. M. The unseen majority: soil microbes as 11 microbiome. We explored various strategies is a central goal in microbial ecology , drivers of plant diversity and productivity in terrestrial of how keystone taxa exert their influence in and it is necessary to extend microbial ecosystems. Ecol. Lett. 11, 296–310 (2008). 5. Graham, E. B. et al. Microbes as engines of ecosystem microbial communities. We also discussed co-occurrence patterns and keystone taxa function: when does community structure enhance how keystone taxa may function individually to ecosystem processes (Fig. 3). Studies predictions of ecosystem processes? Front. Microbiol. 7, 1–10 (2016). or as part of a guild to influence narrow and investigating keystone taxa could include 6. Hamady, M. & Knight, R. Microbial community broad processes. We noted the usefulness promising culturing approaches to explore profiling for human microbiome projects: tools, techniques, and challenges. Genome Res. 19, of correlation scores but emphasized the complex ecological relationships, such as 1141–1152 (2009). importance of causal relationships and commensalism and amensalism, in natural 7. Barberán, A., Bates, S. T., Casamayor, E. O. & Fierer, N. Using network analysis to explore co-occurrence experimental studies for identifying keystone conditions and assess the effect of keystones. patterns in soil microbial communities. ISME J. 6, taxa. To aid future studies, we summarized The actual importance of keystone taxa to 343–351 (2012). 8. Banerjee, S. et al. Determinants of bacterial a number of approaches that can be used microbiome functioning and ecosystem communities in Canadian agroforestry systems. to characterize and harness keystone taxa processes can only be derived from robust Environ. Microbiol. 18, 1805–1816 (2016). 9. Zhang, Z. et al. Spatial heterogeneity and in various ecosystems, and we identified functional profiling using the latest tools, co-occurrence patterns of human mucosal-associated uncharted territories where microbial such as RNA-stable isotope probing107 intestinal microbiota. ISME J. 8, 881–893 (2014). 10. Fuhrman, J. A., Cram, J. A. & Needham, D. M. Marine keystone taxa have not been identified. coupled with metatranscriptomics or microbial community dynamics and their ecological Network scores have been popular to metaproteomics. Upon identifying keystone interpretation. Nat. Rev. Microbiol. 13, 133–146 (2015). statistically identify keystone taxa in recent taxa in an environment, determining if there 11. Prosser, J. I. et al. The role of ecological theory in years, and it is important to continue this are structural keystones and functional microbial ecology. Nat. Rev. Microbiol. 5, 384–392 (2007). momentum to strengthen the repertoire of keystones depending on whether they affect 12. Costello, E. K., Stagaman, K., Dethlefsen, L. & keystone taxa. For a range of taxa, it has been microbiome structure or functioning could Bohannan, B. J. M. & Relman, D. A. The application of ecological theory toward an understanding of the shown that keystone taxa identified using be tested. As any change in microbiome human microbiome. Science 336, 1255–1262 statistical tools indeed have an impact on structure may also have consequences for (2012). 20,39,80 13. Proulx, S. R., Promislow, D. E. L. & Phillips, P. C. microbiome structure and performance . microbiome functioning, a clear distinction Network thinking in ecology and evolution. Trends However, for many other keystone taxa, between structural and functional keystones Ecol. Evol. 20, 345–353 (2005). 14. Zhou, J. et al. Functional molecular ecological such experimental evidence is still missing. in microbial communities is questionable. networks. MBio 1, e00169–00110 (2010). Hence, it is a challenge to complement Nonetheless, the latest molecular tools 15. Reshef, D. N. et al. Detecting novel associations in large data sets. Science 334, 1518–1524 (2011). statistical evidence with empirical evidence have empowered microbiologists to test 16. Ruan, Q. et al. Local similarity analysis reveals unique for keystone taxa in microbial communities such theories and ideas. The contribution associations among marine bacterioplankton species and environmental factors. Bioinformatics 22, (some experimental tools for doing this are of microbial communities for ecosystem 2532–2538 (2006). highlighted above). Moreover, information processes is often missing or insignificant 17. Friedman, J. & Alm, E. J. Inferring correlation on keystone taxa from the desert, tropical in ecosystem models5. These models networks from genomic survey data. PLoS Comput. Biol. 8, 1–11 (2012). forest or vadose zone is rare or not yet mostly consider the overall community 18. Faust, K. et al. Microbial co-occurrence relationships available. Similarly, knowledge of how characteristics (abundance, composition in the human microbiome. PLoS Comput. Biol. 8, e1002606 (2012). keystone taxa respond to environmental and diversity), which might blur the actual 19. Weiss, S. et al. Correlation detection strategies in disturbance, pathogen attack in plants or contribution of important microbial microbial data sets vary widely in sensitivity and precision. ISME J. 10, 1–13 (2016). medical intervention in humans would be members. Keystone taxa observed across 20. Agler, M. T. et al. Microbial hub taxa link host and valuable. For example, determining whether habitats and studies might be the missing abiotic factors to plant microbiome variation. PLoS Biol. 14, 1–31 (2016). keystone taxa help microbiome resilience pieces of the puzzle that could help 21. Gilbert, J. a et al. Defining seasonal marine microbial against perturbations could be tested. The microbial ecologists explain the unexplained community dynamics. ISME J. 6, 298–308 (2012). 22. Fisher, C. K. & Mehta, P. Identifying keystone species role of keystone taxa in plant invasion is an variation in ecosystem processes. in the human gut microbiome from metagenomic equally interesting area, especially in the timeseries using sparse linear regression. PLoS ONE Samiran Banerjee1*, Klaus Schlaeppi 1,2 and 9, 1–10 (2014). light of observations that some invasive Marcel G. A. van der Heijden1,3,4* 23. Shetty, S. A., Hugenholtz, F., Lahti, L., Smidt, H. & de Vos, W. M. Intestinal microbiome landscaping: tree species cannot establish themselves 1 Department of Agroecology and Environment, insight in community assemblage and implications for without their microbial symbionts or that Agroscope, Zurich, Switzerland. microbial modulation strategies. FEMS Microbiol. Rev. 106 41, 182–199 (2017). invasive species alter the soil microbiome . 2Institute of Plant Sciences, University of Bern, Bern, 24. Vick-Majors, T. J., Priscu, J. C. & Amaral-Zettler, L. A. Moreover, our knowledge of fungal, archaeal Switzerland. Modular community structure suggests metabolic and protistan keystone taxa is limited, and 3Department of Evolutionary Biology and plasticity during the transition to polar night in ice- covered Antarctic lakes. ISME J. 8, 778–789 (2014). Environmental Studies, University of Zurich, Zurich, only a few studies have considered fungal– 25. Gokul, J. K. et al. Taxon interactions control the bacteria or fungal–archaeal and bacterial– Switzerland. distributions of cryoconite bacteria colonizing a high 38,39 4Institute of Environmental Biology, Faculty of Science, Arctic ice cap. Mol. Ecol. 25, 3752–3767 (2016). archaeal co-occurrence networks . 26. Comte, J., Lovejoy, C., Crevecoeur, S. & Vincent, W. F. A cross-domain network may reveal how Utrecht University, Utrecht, Netherlands. Co-occurrence patterns in aquatic bacterial *e-mail: [email protected]; communities across changing permafrost landscapes. members of different taxonomic groups Biogeosciences 13, 175–190 (2016). [email protected] associate with each other, whether they share 27. Berry, D. & Widder, S. Deciphering microbial https://doi.org/10.1038/s41579-018-0024-1 interactions and detecting keystone species with resources or whether there are keystone taxa Front. Microbiol. 5 Published online xx xx xxxx co-occurrence networks. , 1–14 important for inter-kingdom associations. (2014).

28. Paine, R. T. Food web complexity and species diversity. 54. Musat, N. et al. A single-cell view on the ecophysiology 80. Herren, C. M. & McMahon, K. D. Small subsets of Am. Nat. 100, 65–75 (1966). of anaerobic phototrophic bacteria. Proc. Natl Acad. highly connected taxa predict compositional change in 29. Mills, L. S. & Doak, D. F. The keystone-species concept Sci. USA 105, 17861–17866 (2008). microbial communities. bioRxiv https://doi.org/ in ecology and conservation. BioScience 43, 219–224 55. Zhao, D. et al. Network analysis reveals seasonal 10.1101/159087 (2017). (1993). variation of co-occurrence correlations between 81. Grace, J. B., Anderson, T., Olff, H. & Scheiner, S. On 30. Cottee-Jones, H. E. W. & Whittaker, R. J. The keystone Cyanobacteria and other bacterioplankton. Sci. Total the specification of structural equation models for species concept: a critical appraisal. Front. Biogeogr. Environ. 573, 817–825 (2016). ecological systems. Ecol. Monogr. 80, 67–87 (2010). 4, 217–220 (2012). 56. Curtis, M. M. et al. The gut commensal bacteroides 82. Lamb, E. G., Shirtliffe, S. J. & May, W. E. Structural 31. Power, M. E. et al. Challenges in the quest for thetaiotaomicron exacerbates enteric infection equation modeling in the plant sciences: an example keystones. Bioscience 46, 609–620 (1996). through modification of the metabolic landscape. using yield components in oat. Can. J. Plant Sci. 91, 32. Paine, R. T. A note on trophic complexity and Cell Host Microbe 16, 759–769 (2014). 603–619 (2011). community stability. Am. Nat. 103, 91–93 (1969). 57. Maldonado-Contreras, A. et al. Structure of the human 83. Banerjee, S., Bora, S., Thrall, P. H. & Richardson, A. E. 33. Dunne, J. A., Williams, R. J. & Martinez, N. D. gastric bacterial community in relation to Helicobacter Soil C and N as causal factors of spatial variation in Network structure and biodiversity loss in food webs: pylori status. ISME J. 5, 574–579 (2011). extracellular enzyme activity across grassland- robustness increase with connectance. Ecol. Lett. 5, 58. Trosvik, P. & de Muinck, E. J. Ecology of bacteria in the woodland ecotones. Appl. Soil Ecol. 105, 1–8 (2016). 558–567 (2002). human gastrointestinal tract — identification of 84. Mouquet, N., Gravel, D., Massol, F. & Calcagno, V. 34. Deng, Y. et al. Molecular ecological network analyses. keystone and foundation taxa. Microbiome 3, 44 Extending the concept of keystone species to BMC Bioinformatics 13, 113 (2012). (2015). communities and ecosystems. Ecol. Lett. 16, 1–8 35. Lupatini, M. et al. Network topology reveals high 59. Wu, S. et al. A human colonic commensal promotes (2013). connectance levels and few key microbial genera colon tumorigenesis via activation of T helper type 17 85. Nichols, D. et al. Use of ichip for high-throughput within soils. Front. Environ. Sci. 2, 1–11 (2014). T cell responses. Nat. Med. 15, 1016–1022 (2009). in situ cultivation of ‘uncultivable’ microbial 36. Zhou, J., Deng, Y., Luo, F., He, Z. & Yang, Y. 60. Ze, X., Duncan, S. H., Louis, P. & Flint, H. J. species. Appl. Environ. Microbiol. 76, 2445–2450 Phylogenetic molecular ecological network of soil Ruminococcus bromii is a keystone species for the (2010). microbial communities in response to elevated CO2. degradation of resistant starch in the human colon. 86. Gavrish, E., Bollmann, A., Epstein, S. & Lewis, K. MBio 2, e00122–e00111 (2011). ISME J. 6, 1535–1543 (2012). A trap for in situ cultivation of filamentous 37. Eldridge, D. J. et al. Soil-foraging animals alter the 61. Ding, J. et al. Soil organic matter quantity and quality actinobacteria. J. Microbiol. Methods 72, 257–262 composition and co-occurrence of microbial communities shape microbial community compositions of (2008). in a desert shrubland. ISME J. 9, 1–11 (2015). subtropical broadleaved forests. Mol. Ecol. 24, 87. Bouguelia, S. et al. On-chip microbial culture for the 38. Ma, B. et al. Geographic patterns of co-occurrence 5175–5185 (2015). specific detection of very low levels of bacteria. Lab. network topological features for soil microbiota at 62. Nunez, M. A. & Dimarco, R. D. The encyclopedia of Chip 13, 4024 (2013). continental scale in eastern China. ISME J. 10, 1–11 sustainability, vol. 5: ecosystem management and 88. Park, J., Kerner, A., Burns, M. A. & Lin, X. N. (2016). sustainability (eds Craig, R. K., Nagle, J. C., Pardy, B., Microdroplet-enabled highly parallel co-cultivation 39. Banerjee, S. et al. Network analysis reveals functional Schmitz, O. J. & Smith, W. K.) 226–230 (Berkshire of microbial communities. PLoS ONE 6, e17019 redundancy and keystone taxa amongst bacterial and Publishing, 2012). (2011). fungal communities during organic matter 63. Hector, A. et al. Plant diversity and productivity 89. Stanley, C. E. & van der Heijden, M. G. A. Microbiome- decomposition in an arable soil. Soil Biol. Biochem. experiments in European grasslands. Science 286, on-a-chip: new frontiers in plant–microbiota research. 97, 188–198 (2016). 1123–1127 (1999). Trends Microbiol. 25, 610–613 (2017). 40. Jiang, Y. et al. Plant cultivars imprint the rhizosphere 64. Hajishengallis, G. et al. Low-abundance biofilm species 90. Scheffer, M., Hosper, S. H., Meijer, M. L., Moss, B. & bacterial community composition and association orchestrates inflammatory periodontal disease Jeppesen, E. Alternative equilibria in shallow lakes. networks. Soil Biol. Biochem. 109, 145–155 (2017). through the commensal microbiota and complement. Trends Ecol. Evol. 8, 275–279 (1993). 41. Li, F., Chen, L., Zhang, J., Yin, J. & Huang, S. Bacterial Cell Host Microbe 10, 497–506 (2011). 91. Schimel, J. P. & Schaeffer, S. M. Microbial control over community structure after long-term organic and 65. Faust, K. & Raes, J. Microbial interactions: from carbon cycling in soil. Front. Microbiol. 3, 1–11 inorganic fertilization reveals important associations networks to models. Nat. Rev. Microbiol. 10, (2012). between soil nutrients and specific taxa involved in 538–550 (2012). 92. Vandenkoornhuyse, P., Quaiser, A., Duhamel, M., nutrient transformations. Front. Microbiol. 8, 187 66. Cardona, C., Weisenhorn, P., Henry, C. & Gilbert, J. A. Le Van, A. & Dufresne, A. The importance of the (2017). Network-based metabolic analysis and microbial microbiome of the plant holobiont. New Phytol. 206, 42. Liang, Y. et al. Long-term oil contamination alters the community modeling. Curr. Opin. Microbiol. 31, 1196–1206 (2015). molecular ecological networks of soil microbial 124–131 (2016). 93. Maloy, K. J. & Powrie, F. Intestinal homeostasis and its functional genes. Front. Microbiol. 7, 1–13 (2016). 67. Strogatz, S. H. Exploring complex networks. Nature breakdown in inflammatory bowel disease. Nature 43. Wang, H. et al. Combined use of network inference 410, 268–276 (2001). 474, 298–306 (2011). tools identifies ecologically meaningful bacterial 68. Newman, M. E. J. The structure and function of 94. Hajishengallis, G. & Lamont, R. J. Dancing with the associations in a paddy soil. Soil Biol. Biochem. 105, complex networks. Soc. Ind. Appl. Math. Rev. 45, stars: how choreographed bacterial interactions 227–235 (2017). 167–256 (2003). dictate nososymbiocity and give rise to keystone 44. Hill, R. et al. Temporal and spatial influences incur 69. van der Heijden, M. G. A. & Hartmann, M. Networking pathogens, accessory pathogens, and pathobionts. reconfiguration of Arctic heathland soil bacterial in the plant microbiome. PLoS Biol. 14, 1–9 (2016). Trends Microbiol. 24, 477–489 (2016). community structure. Environ. Microbiol. 18, 70. Barabási, A. L., Gulbahce, N. & Loscalzo, J. Network 95. Hill, D. & Artis, D. Intestinal bacteria and the 1942–1953 (2016). medicine: a network-based approach to human regulation of immune cell homeostasis. Annu. Rev. 45. Li, B. et al. Metagenomic and network analysis reveal disease. Nat. Rev. Genet. 12, 56–68 (2011). Immunol. 28, 623–667 (2010). wide distribution and co-occurrence of environmental 71. Hartman, K. et al. Cropping practices manipulate 96. Raaijmakers, J. M. & Weller, D. M. Natural plant antibiotic resistance genes. ISME J. 9, 1–13 (2015). abundance patterns of root and soil microbiome protection by 2,4-diacetylphloroglucinol - producing 46. Yang, S. et al. Hydrocarbon degraders establish at the members paving the way to smart farming. Pseudomonas spp. in take-all decline soils. APS. 11, costs of microbial richness, abundance and keystone Microbiome 6, 14 (2018). 144–152 (1998). taxa after crude oil contamination in permafrost 72. Steele, J. A. et al. Marine bacterial, archaeal and 97. van Der Heijden, M. G. A., Martin, F. M., Selosse, M. A. environments. Sci. Rep. 6, 37473 (2016). protistan association networks reveal ecological & Sanders, I. R. Mycorrhizal ecology and evolution: the 47. Chao, Y. et al. Structure, variation, and co-occurrence linkages. ISME J. 5, 1414–1425 (2011). past, the present, and the future. New Phytol. 205, of soil microbial communities in abandoned sites of a 73. Andreote, F. D. et al. Culture-independent assessment 1406–1423 (2015). rare earth elements mine. Environ. Sci. Technol. 50, of rhizobiales-related alphaproteobacteria and the 98. Shade, A. & Handelsman, J. Beyond the Venn 11481–11490 (2016). diversity of Methylobacterium in the rhizosphere and diagram: the hunt for a core microbiome. Environ. 48. Jiao, S. et al. Bacterial communities in oil rhizoplane of transgenic eucalyptus. Microb. Ecol. 57, Microbiol. 14, 4–12 (2012). contaminated soils: biogeography and co-occurrence 82–93 (2009). 99. Turnbaugh, P. J. et al. A core gut microbiom in patterns. Soil Biol. Biochem. 98, 64–73 (2016). 74. Hajishengallis, G., Darveau, R. P. & Curtis, M. A. The obese and lean twins. Nature 457, 480–484 49. Shi, S. et al. The interconnected rhizosphere: high keystone-pathogen hypothesis. Nat. Rev. Microbiol. (2009). network complexity dominates rhizosphere 10, 717–725 (2012). 100. Yeoh, Y. K. et al. Evolutionary conservation of a assemblages. Ecol. Lett. 19, 926–936 (2016). 75. Bäckhed, F., Ley, R. E., Sonnenburg, J. L., Peterson, D. A. core root microbiome across plant phyla along a 50. Yan, Y., Kuramae, E. E., De Hollander, M., & Gordon, J. I. Host-bacterial mutualism in the human tropical soil chronosequence. Nat. Commun. 8, 215 Klinkhamer, P. G. & Van Veen, J. A. Functional traits intestine. Science 307, 1915–1920 (2005). (2017). dominate the diversity-related selection of bacterial 76. Garrett, W. S. et al. Enterobacteriaceae act in concert 101. Brown, J. H. & Heske, E. J. Control of a desert- communities in the rhizosphere. ISME J. 11, 1–11 with the gut microbiota to induce spontaneous and grassland transition by a keystone rodent guild. (2016). maternally transmitted colitis. Cell Host Microbe 8, Science 250, 1705–1707 (1990). 51. Geng, H., Tran-Gyamfi, M. B., Lane, T. W., Sale, K. L. & 292–300 (2010). 102. Jones, C. M. et al. Recently identified microbial guild Yu, E. T. Changes in the structure of the microbial 77. Van Der Heijden, M. G. A. et al. Symbiotic bacteria as mediates soil N2O sink capacity. Nat. Clim. Chang. 4, community associated with Nannochloropsis salina a determinant of plant community structure and plant 801–805 (2014). following treatments with antibiotics and bioactive productivity in dune grassland. FEMS Microbiol. Ecol. 103. Lynch, M. D. J. & Neufeld, J. D. Ecology and compounds. Front. Microbiol. 7, 1–13 (2016). 56, 178–187 (2006). exploration of the rare biosphere. Nat. Rev. Microbiol. 52. Graham, E. B. et al. Deterministic influences exceed 78. Kommineni, S. et al. Bacteriocin production 13, 217–229 (2015). dispersal effects on hydrologically-connected augments niche competition by enterococci in the 104. Jousset, A. et al. Where less may be more: how the microbiomes. Environ. Microbiol. 19, 1552–1567 mammalian gastrointestinal tract. Nature 526, rare biosphere pulls ecosystems strings. ISME J. 11, (2017). 719–722 (2015). 853–862 (2017). 53. Ji, Y. et al. Structure and function of methanogenic 79. van der Heijden, M. G. A. et al. Mycorrhizal fungal 105. Pester, M., Bittner, N., Deevong, P., Wagner, M. & Loy, A. microbial communities in sediments of Amazonian diversity determines plant biodiversity, ecosystem A ‘rare biosphere’ microorganism contributes to lakes with different water types. Environ. Microbiol. variability and productivity. Nature 396, 69–72 sulfate reduction in a peatland. ISME J. 4, 18, 5082–5100 (2016). (1998). 1591–1602 (2010).

106. Stinson, K. A. et al. Invasive plant suppresses the a figure and C. Stanley for proofreading the manuscript. Work Publisher’s note growth of native tree seedlings by disrupting in the author’s laboratory was supported by the Swiss Springer Nature remains neutral with regard to jurisdictional belowground mutualisms. PLoS Biol. 4, 727–731 National Science Foundation (Grant No. 31003A_166079 claims in published maps and institutional affiliations. (2006). awarded to M.G.A.v.d.H.). 107. Manefield, M., Whiteley, A. S., Griffiths, R. I. & Reviewer information Bailey, M. J. RNA stable isotope probing, a novel means Author contributions Nature Reviews Microbiology thanks Janet Jansson and the of linking microbial community function to phylogeny. S.B. researched data for the article. S.B. and M.G.A.v.d.H other anonymous reviewer(s) for their contribution to the Appl. Environ. Microbiol. 68, 5367–5373 (2002). made substantial contributions to the discussion of content peer review of this work. and writing of the article. S.B, K.S. and M.G.A.v.d.H. reviewed Acknowledgements and edited the manuscript before submission. The authors thank the referees, whose constructive com- Supplementary information ments and insightful suggestions greatly improved the quality Competing interests Supplementary information is available for this paper at of the manuscript. They also thank U. Kaufmann for help with The authors declare no competing interests. https://doi.org/10.1038/s41579-018-0024-1.