Deciphering microbial interactions and detecting keystone species with co-occurrence networks.

Berry D, Widder S. · 2014

Read Deciphering microbial interactions and detecting keystone species with co-occurrence networks. by Berry D, Widder S. (2014) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/europmc-pmc4033041.txt.

Introduction

David Berry and Stefanie Widder test whether maps of microbe co-occurrence can reveal true interactions, in Frontiers in Microbiology in 2014. Surveys often list which microbes turn up together, and such patterns are used to guess who helps or hinders whom. The method had rarely been checked against known answers.

The pair built fake microbe communities on a computer with set interaction rules, then sampled them like a real survey. Network maps rebuilt from the samples matched the true interactions only in some settings. When shared habitat taste drove co-occurrence, the maps lost their meaning. Around hub species with many partners, false links clustered thickly. The team also found shape traits in the maps that flag keystone species, the few members a community leans on. Their rules of thumb aim to keep field researchers from over-reading such maps.

Catalog record

CREATORBerry D, Widder S.
YEAR2014
COLLECTIONPAPERS
FORMATTXT
FILE SIZE65 KB
LICENSECC BY 4.0
RIGHTSCC BY
SOURCEEurope PMC

Source: https://europepmc.org/articles/PMC4033041

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Deciphering microbial interactions and detecting keystone species with co-occurrence networks By David Berry, Stefanie Widder Published in: Frontiers in Microbiology DOI: 10.3389/fmicb.2014.00219 ABSTRACT Co-occurrence networks produced from microbial survey sequencing data are frequently used to identify interactions between community members. While this approach has potential to reveal ecological processes, it has been insufficiently validated due to the technical limitations inherent in studying complex microbial ecosystems. Here, we simulate multi-species microbial communities with known interaction patterns using generalized Lotka-Volterra dynamics. We then construct co-occurrence networks and evaluate how well networks reveal the underlying interactions and how experimental and ecological parameters can affect network inference and interpretation. We find that co-occurrence networks can recapitulate interaction networks under certain conditions, but that they lose interpretability when the effects of habitat filtering become significant. We demonstrate that networks suffer from local hot spots of spurious correlation in the neighborhood of hub species that engage in many interactions. We also identify topological features associated with keystone species in co-occurrence networks. This study provides a substantiated framework to guide environmental microbiologists in the construction and interpretation of co-occurrence networks from microbial survey datasets. INTRODUCTION The study of co-occurrence and co-abundance patterns has a long history in ecological research. In macro-ecological surveys non-random species co-occurrence patterns are often observed, indicating that community structure is imprinted by interactions between species. Fundamentally, interactions can be either positive or negative (Faust and Raes, 2012), and consequently lead to either aggregation or alternatively avoidance or exclusion (Cody and Diamond, 1975; Stone and Roberts, 1992; Gotelli and McCabe, 2002). Research on macro-ecological interaction networks and their topologies has revealed that community-wide interaction patterns maximize robustness and functionality (Montoya et al., 2006; Thébault and Fontaine, 2010; Saavedra et al., 2011). They are therefore fundamental units for understanding community dynamics and productivity. Microorganisms also engage in a rich diversity of relationships. Interactions can be antagonistic, such as competition for a limiting resource or direct interference (e.g., by bacteriocin and siderophore production). Interactions can also be cooperative, such as transfer of complementary metabolites (e.g., interspecies H2 transfer) or quorum sensing (Hibbing et al., 2010). Because interactions can affect population dynamics it is expected that signatures of microbial interactions are imprinted in microbial survey datasets. Knowledge about the composition of microbial communities from diverse environments is rapidly expanding due to tremendous advances in sequencing technologies. These technologies power ever broader and deeper surveys using targeted marker genes such as the 16S rRNA gene as well as shotgun metagenomics and metatranscriptomics (Caporaso et al., 2011; Shokralla et al., 2012). Even though the large data sets resulting from these surveys do not provide direct evidence on interaction between species, they are amenable to co-occurrence network construction using correlation coefficients or other association metrics (e.g., Ruan et al., 2006; Barberán et al., 2012; Eiler et al., 2012; Schwab et al., 2014). Ecological interpretations are widely applied to these co-occurrence networks, though validation of these interpretations remains scarce (Faust and Raes, 2012). This lack of validation is particularly problematic in microbial surveys because the physiological capabilities or ecological niches of the organisms under investigation can frequently not even be roughly predicted due to the lack of closely related cultivars or sequenced genomes (Stecher et al., 2013). Most microbial communities are complex, consisting of many species potentially interacting with each other. This makes validation of community-wide interactions difficult if not impossible. In some cases experimental validation can be undertaken by constructing assemblages of organisms and inferring interactions based on combinations of these assemblages (Trosvik et al., 2010). This approach is limited, however, because constructed assemblies may not be relevant to the environment and poorly reflect the complex “wild” ecosystem, and organisms of interest may be so-far uncultured. Alternatively, one can model microbial populations using simple rules about their growth behavior and interaction patterns and thereby simulate dynamics of complex multispecies communities. Mathematical modeling has been used to explore processes such as biofilm formation (Wimpenny and Colasanti, 1997), response of the gut microbiota to changes in diet (Faith et al., 2011), and recovery of the microbial community after antibiotic treatment (Stein et al., 2013). Modeling allows one to investigate a range of situations and parameters and identify under what conditions robust inference can be made. Simulation therefore plays an important role in establishing a framework for exploring microbial diversity and identifying putative interactions in real ecosystems. Here we use generalized Lotka-Volterra (gLV) modeling to simulate the dynamics of multispecies microbial consortia engaged in competitive and cooperative interactions. gLV models are based on simple rules about the effect of each species on each other. They have been applied as representations of microbial communities in diverse environments (Mounier et al., 2008; Rodriguez-Brito et al., 2010; Stein et al., 2013) as well as macro-ecological communities (Case, 2000). By simulating a range of conditions, we evaluate how well co-occurrence networks capture the underlying causal interaction structure by exploring their accuracy for mapping interactions. Investigated conditions cover sampling breadth (i.e., number of samples) and intrinsic ecological parameters such as diversity and interaction structure. We also identify topological features of the network that can be used to predict keystone species. Finally, we provide a summary of best practices and methodological considerations. MATERIALS AND METHODS Simulation of generalized lotka-volterra dynamics Microbial community population dynamics were simulated using a generalized Lotka-Volterra model that follows the form: dxidt=rixi(1−Axki) where for any species i drawn from a meta-community, x is the vector of species abundances, ri is growth rate of species i, ki is the carrying capacity of species i, and A is a matrix containing interaction coefficients between species (Case, 2000). The model was evaluated using numerical integration with the lsoda function in the R package deSolve (Soetaert et al., 2010). Growth rates were assigned to each species from a uniform distribution from >0 to 1 so that all species were capable of positive growth. Carrying capacities were assigned to each species by drawing from either a β distribution, which allows one to simulate a range of distributions from a uniform distribution (with coefficients α = 1, β = 1) to an increasingly uneven distribution (e.g., α = 1 and β > 1), or alternatively from a lognormal distribution to simulate a very uneven distribution. Carrying capacity distributions were scaled to range between 1 and 100. Initial abundances of the species were drawn from a uniform distribution ranging between 10 and 100. The scaling of the carrying capacities and initial abundances was chosen for convenience and do not affect the outcomes of our investigation. The interaction matrix, which determines the topology of the metacommunity network, was assigned from different random models: (i) Erdös-Renyi model (Erdős and Rényi, 1959) resulting in a uniform distribution for interactions, (ii) the Watts-Strogatz model (Watts and Strogatz, 1998) forming networks with small world property, (iii) the Barabasi-Albert model (Barabasi and Albert, 1999) generating scale-free networks, and (iv) the Klemm-Eguiluz model (Klemm and Eguíluz, 2002; Prettejohn et al., 2011) generating topologies with small world and scale-free properties as well as modularity. Interaction magnitude was drawn from a uniform distribution between −1 and 1 (note that a negative value indicates synergy and a positive value antagonism) and the diagonal of the interaction matrix was set to unity to

Read the complete text